statistics

This module contains functions to perform spatial statistics calculations.

P_VALUE

P_VALUE(z_score)

Description

This function computes the p-value (two-tails test) of a given z-score assuming the population follows a normal distribution where the mean is 0 and the standard deviation is 1. The z-score is a measure of how many standard deviations below or above the population mean a value is. It gives you an idea of how far from the mean a data point is. The p-value is the probability that a randomly sampled point has a value at least as extreme as the point whose z-score is being tested.

z_score: FLOAT64

Return type

FLOAT64

Example

SELECT `carto-un`.carto.P_VALUE(u) as p_value
    FROM UNNEST([-2,-1,0,1,2]) u;

-- [ 0.04550012577451279, 0.31731052766472745, 0.999999999, 0.31731052766472745, 0.04550012577451279 ]

SELECT `carto-un-eu`.carto.P_VALUE(u) as p_value
    FROM UNNEST([-2,-1,0,1,2]) u;

-- [ 0.04550012577451279, 0.31731052766472745, 0.999999999, 0.31731052766472745, 0.04550012577451279 ]

SELECT carto.P_VALUE(u) as p_value
    FROM UNNEST([-2,-1,0,1,2]) u;

-- [ 0.04550012577451279, 0.31731052766472745, 0.999999999, 0.31731052766472745, 0.04550012577451279 ]

KNN_TABLE

KNN_TABLE(input, output_table, geoid_col, geo_col, k)

Description

This procedure returns for each point the k-nearest neighbors of a given set of points.

input: STRING the query to the data used to compute the KNN. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
geoid_col: STRING name of the column with unique ids.
geo_col: STRING name of the column with the geometries.
k: INT64 number of nearest neighbors (positive, typically small).

Output

The results are stored in the table named <output_table>, which contains the following columns:

geo: GEOGRAPHY the geometry of the considered point.
geo_knn: GEOGRAPHY the k-nearest neighbor point.
geoid: STRING the unique identifier of the considered point.
geoid_knn: STRING the unique identifier of the k-nearest neighbor.
distance: FLOAT64 the k-nearest neighbor distance to the considered point.
knn: INT64 the k-order (knn).

Example

CALL `carto-un`.carto.KNN_TABLE(
    'myproject.mydataset.points',
    'myproject.mydataset.knn',
    'id',
    'geo',
    10
);

CALL `carto-un-eu`.carto.KNN_TABLE(
    'myproject.mydataset.points',
    'myproject.mydataset.knn',
    'id',
    'geo',
    10
);

CALL carto.KNN_TABLE(
    'myproject.mydataset.points',
    'myproject.mydataset.knn',
    'id',
    'geo',
    10
);

KNN

KNN(points, k)

Description

This function returns for each point the k-nearest neighbors of a given set of points.

points: ARRAY<STRUCT<geoid STRING, geo GEOGRAPHY>> input data with unique id and geography.
k: INT64 number of nearest neighbors (positive, typically small).

Return type

ARRAY<STRUCT<geo GEOGRAPHY, geo_knn GEOGRAPHY, geoid STRING, geoid_knn STRING, distance FLOAT64, knn INT64>>

where:

geo: the geometry of the considered point.
geo_knn: the k-nearest neighbor point.
geoid: the unique identifier of the considered point.
geoid_knn: the unique identifier of the k-nearest neighbor.
distance: the k-nearest neighbor distance to the considered point.
knn: the k-order (knn).

Example

SELECT *
FROM UNNEST((
    SELECT `carto-un`.carto.KNN(myarray, 10)
    FROM (
        SELECT ARRAY_AGG(STRUCT(format('%08x', uid),position_geom)) myarray
        FROM (
            SELECT ROW_NUMBER() OVER (ORDER BY hour) AS uid, position_geom
            FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
            WHERE date = '2017-12-31'
        )
    )
))
ORDER BY geoid;
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.8225 41.97117)",
--  "geoid": "00000001",
--  "geoid_knn": "00000624",
--  "distance": "10.804663098937658",
--  "knn": "1"
--},
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.823 41.9712)",
--  "geoid": "00000001",
--  "geoid_knn": "00000666",
--  "distance": "30.66917920746894",
--  "knn": "2"
--},
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.82298 41.9713)",
--  "geoid": "00000001",
--  "geoid_knn": "00000618",
--  "distance": "31.863463704968353",
--  "knn": "3"
--},
-- ...

SELECT *
FROM UNNEST((
    SELECT `carto-un-eu`.carto.KNN(myarray, 10)
    FROM (
        SELECT ARRAY_AGG(STRUCT(format('%08x', uid),position_geom)) myarray
        FROM (
            SELECT ROW_NUMBER() OVER (ORDER BY hour) AS uid, position_geom
            FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
            WHERE date = '2017-12-31'
        )
    )
))
ORDER BY geoid;
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.8225 41.97117)",
--  "geoid": "00000001",
--  "geoid_knn": "00000624",
--  "distance": "10.804663098937658",
--  "knn": "1"
--},
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.823 41.9712)",
--  "geoid": "00000001",
--  "geoid_knn": "00000666",
--  "distance": "30.66917920746894",
--  "knn": "2"
--},
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.82298 41.9713)",
--  "geoid": "00000001",
--  "geoid_knn": "00000618",
--  "distance": "31.863463704968353",
--  "knn": "3"
--},
-- ...

SELECT *
FROM UNNEST((
    SELECT carto.KNN(myarray, 10)
    FROM (
        SELECT ARRAY_AGG(STRUCT(format('%08x', uid),position_geom)) myarray
        FROM (
            SELECT ROW_NUMBER() OVER (ORDER BY hour) AS uid, position_geom
            FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
            WHERE date = '2017-12-31'
        )
    )
))
ORDER BY geoid;
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.8225 41.97117)",
--  "geoid": "00000001",
--  "geoid_knn": "00000624",
--  "distance": "10.804663098937658",
--  "knn": "1"
--},
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.823 41.9712)",
--  "geoid": "00000001",
--  "geoid_knn": "00000666",
--  "distance": "30.66917920746894",
--  "knn": "2"
--},
--{
--  "geo": "POINT(2.82263 41.97118)",
--  "geo_knn": "POINT(2.82298 41.9713)",
--  "geoid": "00000001",
--  "geoid_knn": "00000618",
--  "distance": "31.863463704968353",
--  "knn": "3"
--},
-- ...

LOF_TABLE

LOF_TABLE(src_fullname STRING, target_fullname STRING, geoid_column_name STRING, geo_column_name STRING, k INT64)

Description

This procedure computes the Local Outlier Factor for each point of a specified column and stores the result in an output table along with the other input columns.

src_fullname: STRING The input table. A STRING of the form project-id.dataset-id.table-name is expected. The project-id can be omitted (in which case the default one will be used).
target_fullname: STRING The resulting table where the LOF will be stored. A STRING of the form project-id.dataset-id.table-name is expected. The project-id can be omitted (in which case the default one will be used). The dataset must exist and the caller needs to have permissions to create a new table in it. The process will fail if the target table already exists.
geoid_column_name: STRING The column name with a unique identifier for each point.
geo_column_name: STRING The column name containing the points.
lof_target_column_name: STRING The column name where the resulting Local Outlier Factor will be stored in the output table.
k: INT64 Number of nearest neighbors (positive, typically small).

Output

The results are stored in the table named <output_table>, which contains the following columns:

geo: GEOGRAPHY the geometry of the considered point.
geoid: GEOGRAPHY the unique identifier of the considered point.
lof: FLOAT64 the Local Outlier Factor score.

Example

CALL `carto-un`.carto.LOF_TABLE(
  'bigquery-public-data.new_york_subway.stations',
  'myproject.mydataset.my_output_table',
  'station_id',
  'station_geom',
  'lof',
  10
);
-- The table `'myproject.mydataset.my_output_table` will be created
-- with an extra column containing the `lof` value.

CALL `carto-un-eu`.carto.LOF_TABLE(
  'bigquery-public-data.new_york_subway.stations',
  'myproject.mydataset.my_output_table',
  'station_id',
  'station_geom',
  'lof',
  10
);
-- The table `'myproject.mydataset.my_output_table` will be created
-- with an extra column containing the `lof` value.

CALL carto.LOF_TABLE(
  'bigquery-public-data.new_york_subway.stations',
  'myproject.mydataset.my_output_table',
  'station_id',
  'station_geom',
  'lof',
  10
);
-- The table `'myproject.mydataset.my_output_table` will be created
-- with an extra column containing the `lof` value.

LOF

LOF(points, k)

Description

This function computes the Local Outlier Factor of each point of a given set of points.

points: ARRAY<STRUCT<geoid STRING, geo GEOGRAPHY>> input data points with unique id and geography.
k: INT64 number of nearest neighbors (positive, typically small).

Return type

ARRAY<STRUCT<geo GEOGRAPHY, geoid GEOGRAPHY, lof FLOAT64>>

where:

geo: the geometry of the considered point.
geoid: the unique identifier of the considered point.
lof: the Local Outlier Factor score.

Example

SELECT *
FROM UNNEST((
    SELECT `carto-un`.carto.LOF(myarray, 10)
    FROM (
        SELECT ARRAY_AGG(STRUCT(format('%08x', uid),position_geom)) myarray
        FROM (
            SELECT ROW_NUMBER() OVER (ORDER BY hour) AS uid, position_geom
            FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
            WHERE date = '2017-12-31'
        )
    )
))
ORDER BY geoid;
-- {"geo": POINT(2.82263 41.97118), "geoid": "00000001", "lof": 1.3217599116891428}
-- {"geo": POINT(2.35705 41.49786), "geoid": "00000002", "lof": 1.235551000737416}
-- {"geo": POINT(2.13967 41.3838), "geoid": "00000003", "lof": 1.1305674032876687}
-- ...

SELECT *
FROM UNNEST((
    SELECT `carto-un-eu`.carto.LOF(myarray, 10)
    FROM (
        SELECT ARRAY_AGG(STRUCT(format('%08x', uid),position_geom)) myarray
        FROM (
            SELECT ROW_NUMBER() OVER (ORDER BY hour) AS uid, position_geom
            FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
            WHERE date = '2017-12-31'
        )
    )
))
ORDER BY geoid;
-- {"geo": POINT(2.82263 41.97118), "geoid": "00000001", "lof": 1.3217599116891428}
-- {"geo": POINT(2.35705 41.49786), "geoid": "00000002", "lof": 1.235551000737416}
-- {"geo": POINT(2.13967 41.3838), "geoid": "00000003", "lof": 1.1305674032876687}
-- ...

SELECT *
FROM UNNEST((
    SELECT carto.LOF(myarray, 10)
    FROM (
        SELECT ARRAY_AGG(STRUCT(format('%08x', uid),position_geom)) myarray
        FROM (
            SELECT ROW_NUMBER() OVER (ORDER BY hour) AS uid, position_geom
            FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
            WHERE date = '2017-12-31'
        )
    )
))
ORDER BY geoid;
-- {"geo": POINT(2.82263 41.97118), "geoid": "00000001", "lof": 1.3217599116891428}
-- {"geo": POINT(2.35705 41.49786), "geoid": "00000002", "lof": 1.235551000737416}
-- {"geo": POINT(2.13967 41.3838), "geoid": "00000003", "lof": 1.1305674032876687}
-- ...

GFUN_TABLE

GFUN_TABLE(input, output_table, geo_col)

Description

This function computes the G-function of a given set of points.

input: STRING the query to the data used to compute the G-Function. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
geo_col: STRING name of the column with the geometries.

Output

The results are stored in the table named <output_table>, which contains the following columns:

distance: FLOAT64 the nearest neighbors distances.
gfun_G: FLOAT64 the empirical G evaluated for each distance in the support.
gfun_ev: FLOAT64 the theoretical Poisson G evaluated for each distance in the support.

Example

CALL `carto-un`.carto.GFUN_TABLE(
    'myproject.mydataset.points',
    'myproject.mydataset.knn',
    'geo'
);

CALL `carto-un-eu`.carto.GFUN_TABLE(
    'myproject.mydataset.points',
    'myproject.mydataset.knn',
    'geo'
);

CALL carto.GFUN_TABLE(
    'myproject.mydataset.points',
    'myproject.mydataset.knn',
    'geo'
);

GFUN

GFUN(points)

Description

This function computes the G-function of a given set of points.

points: ARRAY<GEOGRAPHY> input data points.

Return type

ARRAY<STRUCT<distance FLOAT64, gfun_G FLOAT64, gfun_ev FLOAT64>>

where:

distance: the nearest neighbors distances.
gfun_G: the empirical G evaluated for each distance in the support.
gfun_ev: the theoretical Poisson G evaluated for each distance in the support.

Example

SELECT *
FROM UNNEST((
    SELECT `carto-un`.carto.GFUN(myarray)
    FROM (
        SELECT ARRAY_AGG(position_geom) myarray
        FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
        WHERE date = '2017-12-31'
    )
))
ORDER BY distance;
--{
--  "distance": "38.599968853183",
--  "gfun_G": "0.4319167389418907",
--  "gfun_ev": "4.037383876246414E-4"
--},
--{
--  "distance": "77.199937706366",
--  "gfun_G": "0.5771899392888118",
--  "gfun_ev": "0.0016139757856029613"
--},
--{
--  "distance": "115.799906559549",
--  "gfun_G": "0.6522116218560278",
--  "gfun_ev": "0.003627782844736638"
--},
-- ...

SELECT *
FROM UNNEST((
    SELECT `carto-un-eu`.carto.GFUN(myarray)
    FROM (
        SELECT ARRAY_AGG(position_geom) myarray
        FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
        WHERE date = '2017-12-31'
    )
))
ORDER BY distance;
--{
--  "distance": "38.599968853183",
--  "gfun_G": "0.4319167389418907",
--  "gfun_ev": "4.037383876246414E-4"
--},
--{
--  "distance": "77.199937706366",
--  "gfun_G": "0.5771899392888118",
--  "gfun_ev": "0.0016139757856029613"
--},
--{
--  "distance": "115.799906559549",
--  "gfun_G": "0.6522116218560278",
--  "gfun_ev": "0.003627782844736638"
--},
-- ...

SELECT *
FROM UNNEST((
    SELECT carto.GFUN(myarray)
    FROM (
        SELECT ARRAY_AGG(position_geom) myarray
        FROM `bigquery-public-data.catalonian_mobile_coverage.mobile_data_2015_2017`
        WHERE date = '2017-12-31'
    )
))
ORDER BY distance;
--{
--  "distance": "38.599968853183",
--  "gfun_G": "0.4319167389418907",
--  "gfun_ev": "4.037383876246414E-4"
--},
--{
--  "distance": "77.199937706366",
--  "gfun_G": "0.5771899392888118",
--  "gfun_ev": "0.0016139757856029613"
--},
--{
--  "distance": "115.799906559549",
--  "gfun_G": "0.6522116218560278",
--  "gfun_ev": "0.003627782844736638"
--},
-- ...

CREATE_SPATIAL_COMPOSITE_SUPERVISED

CREATE_SPATIAL_COMPOSITE_SUPERVISED(input_query, index_column, output_prefix, options)

Description

This procedure derives a spatial composite score as the residuals of a regression model which is used to detect areas of under- and over-prediction. The response variable should be measurable and correlated with the set of variables defining the score. For each data point. the residual is defined as the observed value minus the predicted value. Rows with a NULL value in any of the individual variables are dropped.

Input parameters

input_query: STRING the query to the data used to compute the spatial composite. It must contain all the individual variables that should be included in the computation of the composite as well as a unique geographic id for each row. A qualified table name can be given as well, e.g. 'project-id.dataset-id.table-name'.
index_column: STRING the name of the column with the unique geographic identifier.
output_prefix: STRING the prefix for the output table. It should include project and dataset, e.g. 'project-id.dataset-id.table-name'.
options: STRING containing a valid JSON with the different options. Valid options are described below.
- model_transform: STRING containing the TRANSFORM clause in a BigQuery ML CREATE MODEL statement. If NULL no TRANSFORM clause is applied.
- model_options: JSON with the different options allowed by BigQuery ML CREATE MODEL statement for regression models. Any model is allowed as long as it can deal with numerical inputs for the response variable. At least the INPUT_LABEL_COLS and MODEL_TYPE parameters must be specified. By default, data will not be split into train and test (DATA_SPLIT_METHOD = 'NO_SPLIT'). Hyperparameter tuning is not currently supported.
- r2_thr: FLOAT64 the minimum allowed value for the R2 model score. If the R2 of the regression model is lower than this threshold this implies poor fitting and a warning is raised. The default value is 0.5.
- bucketize_method: STRING the method used to discretize the spatial composite score. The default value is NULL. Possible options are:
  - EQUAL_INTERVALS_ZERO_CENTERED: the values of the spatial composite score are discretized into buckets of equal widths centered in zero. The lower and upper limits are derived from the outliers-removed maximum of the absolute values of the score.
- nbuckets: INT64 the number of buckets used when a bucketization method is specified. The default number of buckets is selected using Freedman and Diaconis’s (1981) rule. Ignored if bucketize_method is not specified.
- remove_outliers: BOOL. When bucketize_method is specified, if remove_outliers is set to TRUE the buckets are derived from the oulier-removed data. The outliers are computed using Tukey’s fences k parameter for outlier detection. The default value is TRUE. For large inputs, setting this option to TRUE might cause a Query exceeds CPU resources error. Ignored if bucketize_method is not specified.

Return type

The results are stored in the table named <output_prefix>, which contains the following columns:

index_column: the unique geographic identifier. The type of this column depends on the type of index_column in input_query.
spatial_score: the value of the composite score. The type of this column is FLOAT64 if the score is not discretized and INT64 otherwise.

When the score is discretized by specifying the bucketize_method parameter, the procedure also returns a lookup table named <output_prefix>_lookup_table with the following columns:

lower_bound: FLOAT64 the lower bound of the bin.
upper_bound: FLOAT64 the upper bound of the bin.
spatial_score: INT64 the value of the (discretized) composite score.

Example

CALL `carto-un`.carto.CREATE_SPATIAL_COMPOSITE_SUPERVISED(
    'SELECT * FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "model_transform":[
            "revenue_change",
            "fempop_15_44, public_transport, education, pois, urbanity"
        ],
        "model_options":{
            "MODEL_TYPE":"LINEAR_REG",
            "INPUT_LABEL_COLS":["revenue_change"],
            "DATA_SPLIT_METHOD":"no_split",
            "OPTIMIZE_STRATEGY":"NORMAL_EQUATION",
            "CATEGORY_ENCODING_METHOD":"ONE_HOT_ENCODING"
        },
        "r2_thr":0.4,
        "bucketize_method":"EQUAL_INTERVALS_ZERO_CENTERED",
        "nbuckets":5
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL `carto-un-eu`.carto.CREATE_SPATIAL_COMPOSITE_SUPERVISED(
    'SELECT * FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "model_transform":[
            "revenue_change",
            "fempop_15_44, public_transport, education, pois, urbanity"
        ],
        "model_options":{
            "MODEL_TYPE":"LINEAR_REG",
            "INPUT_LABEL_COLS":["revenue_change"],
            "DATA_SPLIT_METHOD":"no_split",
            "OPTIMIZE_STRATEGY":"NORMAL_EQUATION",
            "CATEGORY_ENCODING_METHOD":"ONE_HOT_ENCODING"
        },
        "r2_thr":0.4,
        "bucketize_method":"EQUAL_INTERVALS_ZERO_CENTERED",
        "nbuckets":5
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL carto.CREATE_SPATIAL_COMPOSITE_SUPERVISED(
    'SELECT * FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "model_transform":[
            "revenue_change",
            "fempop_15_44, public_transport, education, pois, urbanity"
        ],
        "model_options":{
            "MODEL_TYPE":"LINEAR_REG",
            "INPUT_LABEL_COLS":["revenue_change"],
            "DATA_SPLIT_METHOD":"no_split",
            "OPTIMIZE_STRATEGY":"NORMAL_EQUATION",
            "CATEGORY_ENCODING_METHOD":"ONE_HOT_ENCODING"
        },
        "r2_thr":0.4,
        "bucketize_method":"EQUAL_INTERVALS_ZERO_CENTERED",
        "nbuckets":5
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CREATE_SPATIAL_COMPOSITE_UNSUPERVISED

CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(input_query, index_column, output_prefix, options)

Description

This procedure combines (spatial) variables into a meaningful composite score. The composite score can be derived using different methods, scaling and aggregation functions and weights. Rows with a NULL value in any of the model predictors are dropped.

Input parameters

input_query: STRING the query to the data used to compute the spatial composite. It must contain all the individual variables that should be included in the computation of the composite as well as a unique geographic id for each row. A qualified table name can be given as well, e.g. 'project-id.dataset-id.table-name'.
index_column: STRING the name of the column with the unique geographic identifier.
output_prefix: STRING the prefix for the output table. It should include project and dataset, e.g. 'project-id.dataset-id.table-name'.
options: STRING containing a valid JSON with the different options. Valid options are described below. If options is set to NULL then all options are set to default values, as specified in the table below.
- scoring_method: STRING Possible options are ENTROPY, CUSTOM_WEIGHTS, FIRST_PC. With the ENTROPY method the spatial composite is derived as the weighted sum of the proportion of the min-max scaled individual variables, where the weights are based on the entropy of the proportion of each variable. Only numerical variables are allowed. With the CUSTOM_WEIGHTS method, the spatial composite is computed by first scaling each individual variable and then aggregating them according to user-defined scaling and aggregation methods and individual weights. Depending on the scaling parameter, both numerical and ordinal variables are allowed (categorical and boolean variables need to be transformed to ordinal). With the FIRST_PC method, the spatial composite is derived from a Principal Component Analysis as the first principal component score. Only numerical variables are allowed.
- weights: STRUCT the (optional) weights for each variable used to compute the spatial composite when scoring_method is set to CUSTOM_WEIGHTS, passed as {"name":value, …}. If a different scoring method is selected, then this input parameter is ignored. If specified, the sum of the weights must be lower than 1. If no weights are specified, equal weights are assumed. If weights are specified only for some variables and the sum of weights is less than 1, the remainder is distributed equally between the remaining variables. If weights are specified for all the variables and the sum of weights is less than 1, the remainder is distributed equally between all the variables.
- scaling: STRING the user-defined scaling when the scoring_method is set to CUSTOM_WEIGHTS. Possible options are:
  - MIN_MAX_SCALER: data is rescaled into the range [0,1] based on minimum and maximum values. Only numerical variables are allowed.
  - STANDARD_SCALER: data is rescaled by subtracting the mean value and dividing the result by the standard deviation. Only numerical variables are allowed.
  - RANKING: data is replaced by its percent rank, that is by values ranging from 0 lowest to 1. Both numerical and ordinal variables are allowed (categorical and boolean variables need to be transformed to ordinal).
  - DISTANCE_TO_TARGET_MIN(_MAX,_AVG):data is rescaled by dividing by the minimum, maximum, or mean of all the values. Only numerical variables are allowed.
  - PROPORTION: data is rescaled by dividing by the sum total of all the values. Only numerical variables are allowed.
- aggregation: STRING the aggregation function used when the scoring_method is set to CUSTOM_WEIGHTS. Possible options are:
  - LINEAR: the spatial composite is derived as the weighted sum of the scaled individual variables.
  - GEOMETRIC: the spatial composite is given by the product of the scaled individual variables, each to the power of its weight.
- correlation_var: STRING when scoring_method is set to FIRST_PC, the spatial score will be positively correlated with the selected variable (i.e. the sign the spatial score is set such that the correlation between the selected variable and the first principal component score is positive).
- correlation_thr: FLOAT64 the minimum absolute value of the correlation between each individual variable and the first principal component score when scoring_method is set to FIRST_PC.
- return_range: ARRAY<FLOAT64> the user-defined normalization range of the spatial composite score, e.g [0.0,1.0]. Ignored if bucketize_method is specified.
- bucketize_method: STRING the method used to discretize the spatial composite score. Possible options are:
  - EQUAL_INTERVALS: the values of the spatial composite score are discretized into buckets of equal widths.
  - QUANTILES: the values of the spatial composite score are discretized into buckets based on quantiles.
  - JENKS: the values of the spatial composite score are discretized into buckets obtained using k-means clustering.
- nbuckets: INT64 the number of buckets used when a bucketization method is specified. When bucketize_method is set to EQUAL_INTERVALS, if nbuckets is NULL, the default number of buckets is selected using Freedman and Diaconis’s (1981) rule. When bucketize_method is set to JENKS or QUANTILES, nbuckets cannot be NULL. When bucketize_method is set to JENKS the maximum value is 100, aka the maximum number of clusters allowed by BigQuery with k-means clustering.

Option

ENTROPY

CUSTOM_WEIGHTS

FIRST_PC

Valid options

Default value

scoring_method

Optional

ENTROPY, CUSTOM_WEIGHTS, FIRST_PC

ENTROPY

weights

Ignored

Optional

Ignored

{"name":value…}

NULL

scaling

Ignored

Optional

Ignored

MIN_MAX_SCALER, STANDARD_SCALER, RANKING, DISTANCE_TO_TARGET_MIN, DISTANCE_TO_TARGET_MAX, DISTANCE_TO_TARGET_AVG, PROPORTION

MIN_MAX_SCALER

aggregation

Ignored

Optional

Ignored

LINEAR, GEOMETRIC

LINEAR

correlation_var

Ignored

Optional

Mandatory

NULL

correlation_thr

Ignored

Optional

NULL

return_range

Optional

NULL

bucketize_method

Optional

EQUAL_INTERVALS, QUANTILES, JENKS

NULL

nbuckets

Optional

When bucketize_method is set to EQUAL_INTERVALS is selected using Freedman and Diaconis’s (1981) rule

Return type

The results are stored in the table named <output_prefix>, which contains the following columns:

index_column: the unique geographic identifier. The type of this column depends on the type of index_column in input_query.
spatial_score: the value of the composite score. The type of this column is FLOAT64 if the score is not discretized and INT64 otherwise.

When the score is discretized by specifying the bucketize_method parameter, the procedure also returns a lookup table named <output_prefix>_lookup_table with the following columns:

lower_bound: FLOAT64 the lower bound of the bin.
upper_bound: FLOAT64 the upper bound of the bin.
spatial_score: INT64 the value of the (discretized) composite score.

Examples

With the ENTROPY method:

CALL `carto-un`.carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"ENTROPY",
        "return_range":[0.0,1.0]
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL `carto-un-eu`.carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"ENTROPY",
        "return_range":[0.0,1.0]
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"ENTROPY",
        "return_range":[0.0,1.0]
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

With the CUSTOM_WEIGHTS method:

CALL `carto-un`.carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"CUSTOM_WEIGHTS",
        "weights":{"fempop_15_44":0.2,"education":0.1,"urbanity_ordinal":0.1,"pois":0.1},
        "scaling":"RANKING",
        "aggregation":"LINEAR",
        "bucketize_method":"JENKS"
    }
    '''
 )
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL `carto-un-eu`.carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"CUSTOM_WEIGHTS",
        "weights":{"fempop_15_44":0.2,"education":0.1,"urbanity_ordinal":0.1,"pois":0.1},
        "scaling":"RANKING",
        "aggregation":"LINEAR",
        "bucketize_method":"JENKS"
    }
    '''
 )
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"CUSTOM_WEIGHTS",
        "weights":{"fempop_15_44":0.2,"education":0.1,"urbanity_ordinal":0.1,"pois":0.1},
        "scaling":"RANKING",
        "aggregation":"LINEAR",
        "bucketize_method":"JENKS"
    }
    '''
 )
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

With the FIRST_PC method:

CALL `carto-un`.carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"FIRST_PC",
        "correlation_var":"fempop_15_44",
        "correlation_thr":0.6,
        "bucketize_method":"QUANTILES"
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL `carto-un-eu`.carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"FIRST_PC",
        "correlation_var":"fempop_15_44",
        "correlation_thr":0.6,
        "bucketize_method":"QUANTILES"
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL carto.CREATE_SPATIAL_COMPOSITE_UNSUPERVISED(
    'SELECT * EXCEPT(geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    'geoid',
    '<project-id>.<dataset-id>.<table-name>',
    '''{
        "scoring_method":"FIRST_PC",
        "correlation_var":"fempop_15_44",
        "correlation_thr":0.6,
        "bucketize_method":"QUANTILES"
    }
    '''
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CRONBACH_ALPHA_COEFFICIENT

CRONBACH_ALPHA_COEFFICIENT(input_query, output_prefix)

Description

This procedure computes the Cronbach’s alpha coefficient for a set of (spatial) variables. This coefficient can be used as a measure of internal consistency or reliability of the data, based on the strength of correlations between individual variables. Cronbach’s alpha reliability coefficient normally ranges between 0 and 1 but there is actually no lower limit to the coefficient. Higher alpha (closer to 1) vs lower alpha (closer to 0) means higher vs lower consistency, with usually 0.65 being the minimum acceptable value of internal consistency. Rows with a NULL value in any of the individual variables are dropped.

Input parameters

input_query: STRING the query to the data used to compute the coefficient. It must contain all the individual variables that should be included in the computation of the coefficient. A qualified table name can be given as well, e.g. 'project-id.dataset-id.table-name'.
output_prefix: STRING the name for the output table. It should include project and dataset, e.g. 'project-id.dataset-id.table-name'.

Return type

The output table with the following columns:

cronbach_alpha_coef: FLOAT64 the computed Cronbach Alpha coefficient.
k: INT64 the number of the individual variables used to compute the composite.
mean_var: FLOAT64 the mean variance of all individual variables.
mean_cov: FLOAT64 the mean inter-item covariance among all variable pairs.

Example

CALL `carto-un`.carto.CRONBACH_ALPHA_COEFFICIENT(
    'SELECT * EXCEPT(geoid, geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    '<project-id>.<dataset-id>.<table-name>'
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL `carto-un-eu`.carto.CRONBACH_ALPHA_COEFFICIENT(
    'SELECT * EXCEPT(geoid, geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    '<project-id>.<dataset-id>.<table-name>'
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

CALL carto.CRONBACH_ALPHA_COEFFICIENT(
    'SELECT * EXCEPT(geoid, geom, revenue_change, urbanity, urbanity_ordinal) FROM `cartobq.docs.spatial_scoring_input`',
    '<project-id>.<dataset-id>.<table-name>'
)
-- Table `<my-project>.<my-dataset>.<table-name>` will be created

GWR_GRID

GWR_GRID(input_table, features_columns, label_column, cell_column, cell_type, kring_distance, kernel_function, fit_intercept, output_table)

Description

Geographically weighted regression (GWR) models local relationships between spatially varying predictors and an outcome of interest using a local least squares regression.

This procedure performs a local least squares regression for every input cell. This approach was selected to improve computation time and efficiency. The number of models is controlled by the selected cell resolution, thus the user can increase or decrease the resolution of the cell index to perform more or less regressions. Note that you need to provide the cell ID (spatial index) for every location as input (see cell_column parameter), i.e., the cell type and resolution are not passed explicitly, but rather the index has to be computed previously. Hence if you want to increase or decrease the resolution, you need to precompute the corresponding cell ID of every location (see H3 or Quadbin module).

In each regression, the data of the locations in each cell and those of the neighboring cells, defined by the kring_distance parameter, will be taken into account. The data of the neighboring cells will be assigned a lower weight the further they are from the origin cell, following the function specified in the kernel_function. For example, considering cell i and kring_distance of 1. Having n locations located inside cell i, and in the neigheboring cells [n_1, n_2, ..., n_k], then the regression of the cell i will have in total n + n_1 + n_2 + ... + n_k points.

input_table: STRING name of the source dataset. It should be a quoted qualified table with project and dataset: <project-id>.<dataset-id>.<table-name>.
features_columns: ARRAY<STRING> array of column names from input_table to be used as features in the GWR.
label_column: STRING name of the target variable column.
cell_column: STRING name of the column containing the cell ids.
cell_type: STRING spatial index type as 'h3', 'quadbin'.
kring_distance: INT64 distance of the neighboring cells whose data will be included in the local regression of each cell.
kernel_function: STRING kernel function to compute the spatial weights across the kring. Available functions are: 'uniform', 'triangular', 'quadratic', 'quartic' and 'gaussian'.
fit_intercept: BOOL whether to calculate the interception of the model or to force it to zero if, for example, the input data is already supposed to be centered. If NULL, fit_intercept will be considered as TRUE.
output_table: STRING name of the output table. It should be a quoted qualified table with project and dataset: <project-id>.<dataset-id>.<table-name>. The process will fail if the target table already exists. If NULL, the result will be returned directly by the query and not persisted.

Output

The output table will contain a column with the cell id, a column for each feature column containing its corresponding coefficient estimate and one extra column for intercept if fit_intercept is TRUE.

Examples

CALL `carto-un`.carto.GWR_GRID(
    'cartobq.docs.airbnb_berlin_h3_qk_qb',
    ['bedrooms', 'bathrooms'], -- [ beds feature, bathrooms feature ]
    'price', -- price (target variable)
    'h3_z6', 'h3', 3, 'gaussian', TRUE,
    '<project-id>.<dataset-id>.<table-name>'
);

CALL `carto-un-eu`.carto.GWR_GRID(
    'cartobq.docs.airbnb_berlin_h3_qk_qb',
    ['bedrooms', 'bathrooms'], -- [ beds feature, bathrooms feature ]
    'price', -- price (target variable)
    'h3_z6', 'h3', 3, 'gaussian', TRUE,
    '<project-id>.<dataset-id>.<table-name>'
);

CALL carto.GWR_GRID(
    'cartobq.docs.airbnb_berlin_h3_qk_qb',
    ['bedrooms', 'bathrooms'], -- [ beds feature, bathrooms feature ]
    'price', -- price (target variable)
    'h3_z6', 'h3', 3, 'gaussian', TRUE,
    '<project-id>.<dataset-id>.<table-name>'
);

CALL `carto-un`.carto.GWR_GRID(
    'cartobq.docs.airbnb_berlin_h3_qk_qb',
    ['bedrooms', 'bathrooms'], -- [ beds feature, bathrooms feature ]
    'price', -- price (target variable)
    'qb_z12', 'quadbin', 3, 'gaussian', TRUE,
    '<project-id>.<dataset-id>.<table-name>'
);

CALL `carto-un-eu`.carto.GWR_GRID(
    'cartobq.docs.airbnb_berlin_h3_qk_qb',
    ['bedrooms', 'bathrooms'], -- [ beds feature, bathrooms feature ]
    'price', -- price (target variable)
    'qb_z12', 'quadbin', 3, 'gaussian', TRUE,
    '<project-id>.<dataset-id>.<table-name>'
);

CALL carto.GWR_GRID(
    'cartobq.docs.airbnb_berlin_h3_qk_qb',
    ['bedrooms', 'bathrooms'], -- [ beds feature, bathrooms feature ]
    'price', -- price (target variable)
    'qb_z12', 'quadbin', 3, 'gaussian', TRUE,
    '<project-id>.<dataset-id>.<table-name>'
);

Additional examples

Applying GWR to understand Airbnb listings prices

GETIS_ORD_H3_TABLE

GETIS_ORD_H3_TABLE(input, output_table, index_col, value_col, size, kernel)

Description

This procedure computes the Getis-Ord Gi* statistic for each row in the input table.

input: STRING the query to the data used to compute the coefficient. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
index_col: STRING name of the column with the H3 indexes.
value_col: STRING name of the column with the values for each H3 cell.
size: INT64 size of the H3 kring (distance from the origin). This defines the area around each index cell that will be taken into account to compute its Gi* statistic.
kernel: STRING kernel function to compute the spatial weights across the kring. Available functions are: uniform, triangular, quadratic, quartic and gaussian.

The index_col cannot contain NULL values, otherwise a Invalid input origin error will be returned.

Output

The results are stored in the table named <output_table>, which contains the following columns:

index: STRING
gi: FLOAT64 computed Gi* value.
p_value: FLOAT64 computed P value.

Example

CALL `carto-un`.carto.GETIS_ORD_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    3,
    'gaussian'
);

CALL `carto-un-eu`.carto.GETIS_ORD_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    3,
    'gaussian'
);

CALL carto.GETIS_ORD_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    3,
    'gaussian'
);

GETIS_ORD_H3

GETIS_ORD_H3(input, size, kernel)

Description

This function computes the Getis-Ord Gi* statistic for each H3 index in the input array.

input: ARRAY<STRUCT<index STRING, value FLOAT64>> input data with the indexes and values of the cells.
size: INT64 size of the H3 kring (distance from the origin). This defines the area around each index cell that will be taken into account to compute its Gi* statistic.
kernel: STRING kernel function to compute the spatial weights across the kring. Available functions are: uniform, triangular, quadratic, quartic and gaussian.

The input cannot contain NULL indexes values, otherwise a Invalid input origin error will be returned.

Return type

ARRAY<STRUCT<index STRING, gi FLOAT64, p_value FLOAT64>>

Example

SELECT `carto-un`.carto.GETIS_ORD_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('89394460c37ffff', 28.0),
        STRUCT('89394460077ffff', 19.0)
    ],
    3, 'gaussian'
);
-- {"index": "89394460323ffff", "gi": 1.3606194139870573, "p_value": 0.17363411613079893}
-- {"index": "89394460c37ffff", "gi": -0.34633948719670526, "p_value": 0.7290877280096945}
-- {"index": "89394460077ffff", "gi": -1.0142799267903515, "p_value": 0.31044923023489734}

SELECT `carto-un-eu`.carto.GETIS_ORD_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('89394460c37ffff', 28.0),
        STRUCT('89394460077ffff', 19.0)
    ],
    3, 'gaussian'
);
-- {"index": "89394460323ffff", "gi": 1.3606194139870573, "p_value": 0.17363411613079893}
-- {"index": "89394460c37ffff", "gi": -0.34633948719670526, "p_value": 0.7290877280096945}
-- {"index": "89394460077ffff", "gi": -1.0142799267903515, "p_value": 0.31044923023489734}

SELECT carto.GETIS_ORD_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('89394460c37ffff', 28.0),
        STRUCT('89394460077ffff', 19.0)
    ],
    3, 'gaussian'
);
-- {"index": "89394460323ffff", "gi": 1.3606194139870573, "p_value": 0.17363411613079893}
-- {"index": "89394460c37ffff", "gi": -0.34633948719670526, "p_value": 0.7290877280096945}
-- {"index": "89394460077ffff", "gi": -1.0142799267903515, "p_value": 0.31044923023489734}

SELECT `carto-un`.carto.GETIS_ORD_H3(input_data, 3, 'gaussian')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

SELECT `carto-un-eu`.carto.GETIS_ORD_H3(input_data, 3, 'gaussian')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

SELECT carto.GETIS_ORD_H3(input_data, 3, 'gaussian')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

Additional examples

Identifying amenity hotspots in Stockholm

GETIS_ORD_QUADBIN_TABLE

GETIS_ORD_QUADBIN_TABLE(input, output_table, index_col, value_col, size, kernel)

Description

This procedure computes the Getis-Ord Gi* statistic for each row in the input table.

input: STRING the query to the data used to compute the coefficient. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>
index_col: STRING name of the column with the Quadbin indexes.
value_col: STRING name of the column with the values for each Quadbin cell.
size: INT64 size of the Quadbin kring (distance from the origin). This defines the area around each index cell that will be taken into account to compute its Gi* statistic.
kernel: STRING kernel function to compute the spatial weights across the kring. Available functions are: uniform, triangular, quadratic, quartic and gaussian.

The index_col cannot contain NULL values, otherwise a Invalid input origin error will be returned.

Output

The results are stored in the table named <output_table>, which contains the following columns:

index: INT64
gi: FLOAT64 computed Gi* value.
p_value: FLOAT64 computed P value.

Example

CALL `carto-un`.carto.GETIS_ORD_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    3,
    'gaussian'
);

CALL `carto-un-eu`.carto.GETIS_ORD_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    3,
    'gaussian'
);

CALL carto.GETIS_ORD_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    3,
    'gaussian'
);

GETIS_ORD_QUADBIN

GETIS_ORD_QUADBIN(input, size, kernel)

Description

This function computes the Getis-Ord Gi* statistic for each Quadbin index in the input array.

input: ARRAY<STRUCT<index INT64, value FLOAT64>> input data with the indexes and values of the cells.
size: INT64 size of the Quadbin k-ring (distance from the origin). This defines the area around each index cell that will be taken into account to compute its Gi* statistic.
kernel: STRING kernel function to compute the spatial weights across the kring. Available functions are: uniform, triangular, quadratic, quartic and gaussian.

The input cannot contain NULL indexes values, otherwise a Invalid input origin error will be returned.

Return type

ARRAY<STRUCT<index INT64, gi FLOAT64, p_value FLOAT64>>

Example

SELECT `carto-un`.carto.GETIS_ORD_QUADBIN(
    [
        STRUCT(5266443791933898751, 51.0),
        STRUCT(5266443803500740607, 28.0),
        STRUCT(5266443790415822847, 19.0)
    ],
    3, 'gaussian'
);
-- {"index": 5266443791933898751, "gi": 1.360619413987058, "p_value": 0.0.17363411613079893}
-- {"index": 5266443803500740607, "gi": -0.3463394871967051, "p_value": 0.0.7290877280096945}
-- {"index": 5266443790415822847, "gi": -1.0142799267903515, "p_value": 0.31044923023489734}

SELECT `carto-un-eu`.carto.GETIS_ORD_QUADBIN(
    [
        STRUCT(5266443791933898751, 51.0),
        STRUCT(5266443803500740607, 28.0),
        STRUCT(5266443790415822847, 19.0)
    ],
    3, 'gaussian'
);
-- {"index": 5266443791933898751, "gi": 1.360619413987058, "p_value": 0.0.17363411613079893}
-- {"index": 5266443803500740607, "gi": -0.3463394871967051, "p_value": 0.0.7290877280096945}
-- {"index": 5266443790415822847, "gi": -1.0142799267903515, "p_value": 0.31044923023489734}

SELECT carto.GETIS_ORD_QUADBIN(
    [
        STRUCT(5266443791933898751, 51.0),
        STRUCT(5266443803500740607, 28.0),
        STRUCT(5266443790415822847, 19.0)
    ],
    3, 'gaussian'
);
-- {"index": 5266443791933898751, "gi": 1.360619413987058, "p_value": 0.0.17363411613079893}
-- {"index": 5266443803500740607, "gi": -0.3463394871967051, "p_value": 0.0.7290877280096945}
-- {"index": 5266443790415822847, "gi": -1.0142799267903515, "p_value": 0.31044923023489734}

SELECT `carto-un`.carto.GETIS_ORD_QUADBIN(input_data, 3, 'gaussian')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

SELECT `carto-un-eu`.carto.GETIS_ORD_QUADBIN(input_data, 3, 'gaussian')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

SELECT carto.GETIS_ORD_QUADBIN(input_data, 3, 'gaussian')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

GETIS_ORD_SPACETIME_H3_TABLE

GETIS_ORD_SPACETIME_H3_TABLE(input, output_table, index_col, date_col, value_col, size, time_freq, time_bw, kernel, kernel_time)

Description

This procedure computes the space temporal Getis-Ord Gi* statistic for each H3 index and each datetime timestamp according to the method described in this paper. It extends the Getis-Ord Gi* function by including the time domain. The Getis-Ord Gi* statistic is a measure of spatial autocorrelation, which is the degree to which data values are clustered together in space and time. The statistic is computed as the sum of the values of the cells in the kring (distance from the origin, space and temporal) weighted by the kernel functions, minus the value of the origin cell, divided by the standard deviation of the values of the cells in the kring. The Getis-Ord Gi* statistic is calculated from minimum to maximum datetime with the step defined by the user, in the input array. The datetime timestamp is truncated to the provided level, for example day / hour / week etc. For each spatial index, the missing datetime timestamp, from minimum to maximum, are filled with the default value of 0. Any other imputation of the values should take place outside of the function prior to passing the input to the function. The p value is computed as the probability of observing a value as extreme as the observed value, assuming the null hypothesis that the values are randomly distributed in space and time. The p value is computed using a normal distribution approximation.

input: STRING the query to the data used to compute the coefficient. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
index_col: STRING name of the column with the H3 indexes.
date_col: STRING name of the column with the date.
value_col: STRING name of the column with the values for each H3 cell.
size: INT64 size of the H3 kring (distance from the origin). This defines the area around each index cell that will be taken into account to compute its Gi* statistic.
time_freq: STRING The time interval - step to use for the time series. Available values are: year, quarter, month, week, day, hour, minute, second. It is the equivalent of the spatial index in the time domain.
time_bw: INT64 The bandwidth to use for the time series. This defines the number of adjacent observations in time domain to be considered. It is the equivalent of the H3 kring in the time domain.
kernel: STRING kernel function to compute the spatial weights across the kring. Available functions are: uniform, triangular, quadratic, quartic and gaussian.
kernel_time: STRING kernel function to compute the temporal weights within the time bandwidth. Available functions are: uniform, triangular, quadratic, quartic and gaussian.

The index_col cannot contain NULL values, otherwise a Invalid input origin error will be returned.

Output

The results are stored in the table named <output_table>, which contains the following columns:

index: STRING
date: DATETIME
gi: FLOAT64 computed Gi* value.
p_value: FLOAT64 computed P value.

Example

CALL `carto-un`.carto.GETIS_ORD_SPACETIME_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'date',
    'value',
    3,
    'DAY',
    1,
    'gaussian',
    'gaussian'
);

CALL `carto-un-eu`.carto.GETIS_ORD_SPACETIME_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'date',
    'value',
    3,
    'DAY',
    1,
    'gaussian',
    'gaussian'
);

CALL carto.GETIS_ORD_SPACETIME_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'date',
    'value',
    3,
    'DAY',
    1,
    'gaussian',
    'gaussian'
);

GETIS_ORD_SPACETIME_H3

GETIS_ORD_SPACETIME_H3(input, size, time_freq, time_bw, kernel, kernel_time)

Description

This table function computes the space temporal Getis-Ord Gi* statistic for each H3 index and each datetime timestamp according to the method described in this paper. It extends the Getis-Ord Gi* function by including the time domain. The Getis-Ord Gi* statistic is a measure of spatial autocorrelation, which is the degree to which data values are clustered together in space and time. The statistic is computed as the sum of the values of the cells in the kring (distance from the origin, space and temporal) weighted by the kernel functions, minus the value of the origin cell, divided by the standard deviation of the values of the cells in the kring. The Getis-Ord Gi* statistic is calculated from minimum to maximum datetime with the step defined by the user, in the input array. The datetime timestamp is truncated to the provided level, for example day / hour / week etc. For each spatial index, the missing datetime timestamp, from minimum to maximum, are filled with the default value of 0. Any other imputation of the values should take place outside of the function prior to passing the input to the function. The p value is computed as the probability of observing a value as extreme as the observed value, assuming the null hypothesis that the values are randomly distributed in space and time. The p value is computed using a normal distribution approximation.

input: ARRAY<STRUCT<index STRING, date DATETIME, value FLOAT64>> input data with the indexes and values of the cells.
size: INT64 size of the H3 kring (distance from the origin). This defines the area around each index cell that will be taken into account to compute its Gi* statistic.
time_freq: STRING The time interval - step to use for the time series. Available values are: year, quarter, month, week, day, hour, minute, second. It is the equivalent of the spatial index in the time domain.
time_bw: INT64 The bandwidth to use for the time series. This defines the number of adjacent observations in time domain to be considered. It is the equivalent of the H3 kring in the time domain.
kernel: STRING kernel function to compute the spatial weights across the kring. Available functions are: uniform, triangular, quadratic, quartic and gaussian.
kernel_time: STRING kernel function to compute the temporal weights within the time bandwidth. Available functions are: uniform, triangular, quadratic, quartic and gaussian.

The input cannot contain NULL indexes values, otherwise a Invalid input origin error will be returned.

Return type

TABLE(index STRING, date DATETIME, gi FLOAT64, p_value FLOAT64)

Example

SELECT * FROM `carto-un`.carto.GETIS_ORD_SPACETIME_H3(
    [
        STRUCT('89394460323ffff', DATETIME(2023, 5, 1, 0, 0, 0), 51.0),
        STRUCT('89394460c37ffff', DATETIME(2023, 5, 2, 0, 0, 0), 28.0),
        STRUCT('89394460077ffff', DATETIME(2023, 5, 3, 0, 0, 0), 19.0)
    ],
    3, 'DAY', 1, 'gaussian', 'gaussian'
);
-- index, date, gi, p_value
-- 89394460077ffff 2023-05-01 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 89394460077ffff 2023-05-02 00:00:00.000 -0.6294621529087477 0.5290464242343208
-- 89394460077ffff 2023-05-03 00:00:00.000 0.07918630608303281 0.9368843022965981
-- 89394460323ffff 2023-05-01 00:00:00.000 1.7610199325971272 0.07823494063450298
-- 89394460323ffff 2023-05-02 00:00:00.000 0.3421868170238157 0.732210438784662
-- 89394460323ffff 2023-05-03 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 89394460c37ffff 2023-05-01 00:00:00.000 -0.026829208020286936 0.9785958577333086
-- 89394460c37ffff 2023-05-02 00:00:00.000 0.1953523050277685 0.8451171948001063
-- 89394460c37ffff 2023-05-03 00:00:00.000 -0.026829208020286936 0.9785958577333086

SELECT * FROM `carto-un-eu`.carto.GETIS_ORD_SPACETIME_H3(
    [
        STRUCT('89394460323ffff', DATETIME(2023, 5, 1, 0, 0, 0), 51.0),
        STRUCT('89394460c37ffff', DATETIME(2023, 5, 2, 0, 0, 0), 28.0),
        STRUCT('89394460077ffff', DATETIME(2023, 5, 3, 0, 0, 0), 19.0)
    ],
    3, 'DAY', 1, 'gaussian', 'gaussian'
);
-- index, date, gi, p_value
-- 89394460077ffff 2023-05-01 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 89394460077ffff 2023-05-02 00:00:00.000 -0.6294621529087477 0.5290464242343208
-- 89394460077ffff 2023-05-03 00:00:00.000 0.07918630608303281 0.9368843022965981
-- 89394460323ffff 2023-05-01 00:00:00.000 1.7610199325971272 0.07823494063450298
-- 89394460323ffff 2023-05-02 00:00:00.000 0.3421868170238157 0.732210438784662
-- 89394460323ffff 2023-05-03 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 89394460c37ffff 2023-05-01 00:00:00.000 -0.026829208020286936 0.9785958577333086
-- 89394460c37ffff 2023-05-02 00:00:00.000 0.1953523050277685 0.8451171948001063
-- 89394460c37ffff 2023-05-03 00:00:00.000 -0.026829208020286936 0.9785958577333086

SELECT * FROM carto.GETIS_ORD_SPACETIME_H3(
    [
        STRUCT('89394460323ffff', DATETIME(2023, 5, 1, 0, 0, 0), 51.0),
        STRUCT('89394460c37ffff', DATETIME(2023, 5, 2, 0, 0, 0), 28.0),
        STRUCT('89394460077ffff', DATETIME(2023, 5, 3, 0, 0, 0), 19.0)
    ],
    3, 'DAY', 1, 'gaussian', 'gaussian'
);
-- index, date, gi, p_value
-- 89394460077ffff 2023-05-01 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 89394460077ffff 2023-05-02 00:00:00.000 -0.6294621529087477 0.5290464242343208
-- 89394460077ffff 2023-05-03 00:00:00.000 0.07918630608303281 0.9368843022965981
-- 89394460323ffff 2023-05-01 00:00:00.000 1.7610199325971272 0.07823494063450298
-- 89394460323ffff 2023-05-02 00:00:00.000 0.3421868170238157 0.732210438784662
-- 89394460323ffff 2023-05-03 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 89394460c37ffff 2023-05-01 00:00:00.000 -0.026829208020286936 0.9785958577333086
-- 89394460c37ffff 2023-05-02 00:00:00.000 0.1953523050277685 0.8451171948001063
-- 89394460c37ffff 2023-05-03 00:00:00.000 -0.026829208020286936 0.9785958577333086

SELECT * FROM `carto-un`.carto.GETIS_ORD_SPACETIME_H3((
    SELECT ARRAY_AGG(STRUCT(index, timestamp, value)) AS input_data
    FROM mytable
), 3, 'DAY', 1, 'gaussian', 'gaussian');

SELECT * FROM `carto-un-eu`.carto.GETIS_ORD_SPACETIME_H3((
    SELECT ARRAY_AGG(STRUCT(index, timestamp, value)) AS input_data
    FROM mytable
), 3, 'DAY', 1, 'gaussian', 'gaussian');

SELECT * FROM carto.GETIS_ORD_SPACETIME_H3((
    SELECT ARRAY_AGG(STRUCT(index, timestamp, value)) AS input_data
    FROM mytable
), 3, 'DAY', 1, 'gaussian', 'gaussian');

GETIS_ORD_SPACETIME_QUADBIN_TABLE

GETIS_ORD_SPACETIME_QUADBIN_TABLE(input, output_table, index_col, date_col, value_col, size, time_freq, time_bw, kernel, kernel_time)

Description

This procedure computes the space temporal Getis-Ord Gi* statistic for each Quadbin index and each datetime timestamp according to the method described in this paper. It extends the Getis-Ord Gi* function by including the time domain. The Getis-Ord Gi* statistic is a measure of spatial autocorrelation, which is the degree to which data values are clustered together in space and time. The statistic is computed as the sum of the values of the cells in the kring (distance from the origin, space and temporal) weighted by the kernel functions, minus the value of the origin cell, divided by the standard deviation of the values of the cells in the kring. The Getis-Ord Gi* statistic is calculated from minimum to maximum datetime with the step defined by the user, in the input array. The datetime timestamp is truncated to the provided level, for example day / hour / week etc. For each spatial index, the missing datetime timestamp, from minimum to maximum, are filled with the default value of 0. Any other imputation of the values should take place outside of the function prior to passing the input to the function. The p value is computed as the probability of observing a value as extreme as the observed value, assuming the null hypothesis that the values are randomly distributed in space and time. The p value is computed using a normal distribution approximation.

input: STRING the query to the data used to compute the coefficient. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
index_col: STRING name of the column with the Quadbin indexes.
date_col: STRING name of the column with the date.
value_col: STRING name of the column with the values for each Quadbin cell.
size: INT64 size of the Quadbin kring (distance from the origin). This defines the area around each index cell that will be taken into account to compute its Gi* statistic.
time_freq: STRING The time interval - step to use for the time series. Available values are: year, quarter, month, week, day, hour, minute, second. It is the equivalent of the spatial index in the time domain.
time_bw: INT64 The bandwidth to use for the time series. This defines the number of adjacent observations in time domain to be considered. It is the equivalent of the Quadbin kring in the time domain.
kernel: STRING kernel function to compute the spatial weights across the kring. Available functions are: uniform, triangular, quadratic, quartic and gaussian.
kernel_time: STRING kernel function to compute the temporal weights within the time bandwidth. Available functions are: uniform, triangular, quadratic, quartic and gaussian.

The index_col cannot contain NULL values, otherwise a Invalid input origin error will be returned.

Output

The results are stored in the table named <output_table>, which contains the following columns:

index: INT64
date: DATETIME
gi: FLOAT64 computed Gi* value.
p_value: FLOAT64 computed P value.

Example

CALL `carto-un`.carto.GETIS_ORD_SPACETIME_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'date',
    'value'.
    3,
    'DAY',
    1,
    'gaussian',
    'gaussian'
);

CALL `carto-un-eu`.carto.GETIS_ORD_SPACETIME_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'date',
    'value'.
    3,
    'DAY',
    1,
    'gaussian',
    'gaussian'
);

CALL carto.GETIS_ORD_SPACETIME_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'date',
    'value'.
    3,
    'DAY',
    1,
    'gaussian',
    'gaussian'
);

GETIS_ORD_SPACETIME_QUADBIN

GETIS_ORD_SPACETIME_QUADBIN(input, size, time_freq, time_bw, kernel, kernel_time)

Description

This table function computes the space temporal Getis-Ord Gi* statistic for each Quadbin index and each datetime timestamp according to the method described in this paper. It extends the Getis-Ord Gi* function by including the time domain. The Getis-Ord Gi* statistic is a measure of spatial autocorrelation, which is the degree to which data values are clustered together in space and time. The statistic is computed as the sum of the values of the cells in the kring (distance from the origin, space and temporal) weighted by the kernel functions, minus the value of the origin cell, divided by the standard deviation of the values of the cells in the kring. The Getis-Ord Gi* statistic is calculated from minimum to maximum datetime with the step defined by the user, in the input array. The datetime timestamp is truncated to the provided level, for example day / hour / week etc. For each spatial index, the missing datetime timestamp, from minimum to maximum, are filled with the default value of 0. Any other imputation of the values should take place outside of the function prior to passing the input to the function. The p value is computed as the probability of observing a value as extreme as the observed value, assuming the null hypothesis that the values are randomly distributed in space and time. The p value is computed using a normal distribution approximation.

input: ARRAY<STRUCT<index INT64, date DATETIME, value FLOAT64>> input data with the indexes and values of the cells.
size: INT64 size of the Quadbin kring (distance from the origin). This defines the area around each index cell that will be taken into account to compute its Gi* statistic.
time_freq: STRING The time interval - step to use for the time series. Available values are: year, quarter, month, week, day, hour, minute, second. It is the equivalent of the spatial index in the time domain.
time_bw: INT64 The bandwidth to use for the time series. This defines the number of adjacent observations in time domain to be considered. It is the equivalent of the Quadbin kring in the time domain.
kernel: STRING kernel function to compute the spatial weights across the kring. Available functions are: uniform, triangular, quadratic, quartic and gaussian.
kernel_time: STRING kernel function to compute the temporal weights within the time bandwidth. Available functions are: uniform, triangular, quadratic, quartic and gaussian.

The input cannot contain NULL indexes values, otherwise a Invalid input origin error will be returned.

Return type

TABLE(index INT64, date DATETIME, gi FLOAT64, p_value FLOAT64)

Example

SELECT * FROM `carto-un`.carto.GETIS_ORD_SPACETIME_QUADBIN(
    [
        STRUCT(5266443791933898751, DATETIME(2023, 5, 1, 0, 0, 0), 51.0),
        STRUCT(5266443803500740607, DATETIME(2023, 5, 2, 0, 0, 0), 28.0),
        STRUCT(5266443790415822847, DATETIME(2023, 5, 3, 0, 0, 0), 19.0)
    ],
    3, 'DAY', 1, 'gaussian', 'gaussian'
);
-- index, date, gi, p_value
-- 5266443790415822847 2023-05-01 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 5266443790415822847 2023-05-02 00:00:00.000 -0.6294621529087477 0.5290464242343208
-- 5266443790415822847 2023-05-03 00:00:00.000 0.07918630608303281 0.9368843022965981
-- 5266443791933898751 2023-05-01 00:00:00.000 1.7610199325971272 0.07823494063450298
-- 5266443791933898751 2023-05-02 00:00:00.000 0.3421868170238157 0.732210438784662
-- 5266443791933898751 2023-05-03 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 5266443803500740607 2023-05-01 00:00:00.000 -0.026829208020286936 0.9785958577333086
-- 5266443803500740607 2023-05-02 00:00:00.000 0.1953523050277685 0.8451171948001063
-- 5266443803500740607 2023-05-03 00:00:00.000 -0.026829208020286936 0.9785958577333086

SELECT * FROM `carto-un-eu`.carto.GETIS_ORD_SPACETIME_QUADBIN(
    [
        STRUCT(5266443791933898751, DATETIME(2023, 5, 1, 0, 0, 0), 51.0),
        STRUCT(5266443803500740607, DATETIME(2023, 5, 2, 0, 0, 0), 28.0),
        STRUCT(5266443790415822847, DATETIME(2023, 5, 3, 0, 0, 0), 19.0)
    ],
    3, 'DAY', 1, 'gaussian', 'gaussian'
);
-- index, date, gi, p_value
-- 5266443790415822847 2023-05-01 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 5266443790415822847 2023-05-02 00:00:00.000 -0.6294621529087477 0.5290464242343208
-- 5266443790415822847 2023-05-03 00:00:00.000 0.07918630608303281 0.9368843022965981
-- 5266443791933898751 2023-05-01 00:00:00.000 1.7610199325971272 0.07823494063450298
-- 5266443791933898751 2023-05-02 00:00:00.000 0.3421868170238157 0.732210438784662
-- 5266443791933898751 2023-05-03 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 5266443803500740607 2023-05-01 00:00:00.000 -0.026829208020286936 0.9785958577333086
-- 5266443803500740607 2023-05-02 00:00:00.000 0.1953523050277685 0.8451171948001063
-- 5266443803500740607 2023-05-03 00:00:00.000 -0.026829208020286936 0.9785958577333086

SELECT * FROM carto.GETIS_ORD_SPACETIME_QUADBIN(
    [
        STRUCT(5266443791933898751, DATETIME(2023, 5, 1, 0, 0, 0), 51.0),
        STRUCT(5266443803500740607, DATETIME(2023, 5, 2, 0, 0, 0), 28.0),
        STRUCT(5266443790415822847, DATETIME(2023, 5, 3, 0, 0, 0), 19.0)
    ],
    3, 'DAY', 1, 'gaussian', 'gaussian'
);
-- index, date, gi, p_value
-- 5266443790415822847 2023-05-01 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 5266443790415822847 2023-05-02 00:00:00.000 -0.6294621529087477 0.5290464242343208
-- 5266443790415822847 2023-05-03 00:00:00.000 0.07918630608303281 0.9368843022965981
-- 5266443791933898751 2023-05-01 00:00:00.000 1.7610199325971272 0.07823494063450298
-- 5266443791933898751 2023-05-02 00:00:00.000 0.3421868170238157 0.732210438784662
-- 5266443791933898751 2023-05-03 00:00:00.000 -0.9194024096597111 0.3578850857359205
-- 5266443803500740607 2023-05-01 00:00:00.000 -0.026829208020286936 0.9785958577333086
-- 5266443803500740607 2023-05-02 00:00:00.000 0.1953523050277685 0.8451171948001063
-- 5266443803500740607 2023-05-03 00:00:00.000 -0.026829208020286936 0.9785958577333086

SELECT * FROM `carto-un`.carto.GETIS_ORD_SPACETIME_QUADBIN((
    SELECT ARRAY_AGG(STRUCT(index, timestamp, value)) AS input_data
    FROM mytable
), 3, 'DAY', 1, 'gaussian', 'gaussian');

SELECT * FROM `carto-un-eu`.carto.GETIS_ORD_SPACETIME_QUADBIN((
    SELECT ARRAY_AGG(STRUCT(index, timestamp, value)) AS input_data
    FROM mytable
), 3, 'DAY', 1, 'gaussian', 'gaussian');

SELECT * FROM carto.GETIS_ORD_SPACETIME_QUADBIN((
    SELECT ARRAY_AGG(STRUCT(index, timestamp, value)) AS input_data
    FROM mytable
), 3, 'DAY', 1, 'gaussian', 'gaussian');

MORANS_I_H3_TABLE

MORANS_I_H3_TABLE(input, output_table, index_col, value_col, size, decay)

Description

This procedure computes the Moran's I spatial autocorrelation from the input table with H3 indexes.

input: STRING the query to the data used to compute the coefficient. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
index_col: STRING name of the column with the H3 indexes.
value_col: STRING name of the column with the values for each H3 cell.
size: INT64 size of the H3 k-ring (distance from the origin). This defines the area around each index cell where the distance decay will be applied. If no neighboring cells are found, the weight of the corresponding index cell is set to zero.
decay: STRING decay function to compute the distance decay. Available functions are: uniform, inverse, inverse_square and exponential.

The index_col cannot contain NULL values, otherwise a Invalid input origin error will be returned.

Output

The results are stored in the table named <output_table>, which contains the following column:

morans_i: FLOAT64 Moran's I spatial autocorrelation.

If all cells have no neighbours, then the procedure will fail.

Example

CALL `carto-un`.carto.MORANS_I_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    5,
    'uniform'
);

CALL `carto-un-eu`.carto.MORANS_I_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    5,
    'uniform'
);

CALL carto.MORANS_I_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    5,
    'uniform'
);

MORANS_I_H3

MORANS_I_H3(input, size, decay)

Description

This function computes the Moran's I spatial autocorrelation from the input array of H3 indexes.

input: ARRAY<STRUCT<index STRING, value FLOAT64>> input data with the indexes and values of the cells.
size: INT64 size of the H3 k-ring (distance from the origin). This defines the area around each index cell where the distance decay will be applied. If the cells don't have neighbours given the kring size NULL is returned. If no neighboring cells are found, the weight of the corresponding index cell is set to zero.
decay: STRING decay function to compute the distance decay. Available functions are: uniform, inverse, inverse_square and exponential.

The input cannot contain NULL indexes values, otherwise a Invalid input origin error will be returned.

Return type

FLOAT64. If all cells have no neighbours, then the function will fail.

Example

SELECT `carto-un`.carto.MORANS_I_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('89394460c37ffff', 28.0),
        STRUCT('89394460077ffff', 19.0)
    ],
    8, 'exponential'
);
-- -0.92003263973888194

SELECT `carto-un-eu`.carto.MORANS_I_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('89394460c37ffff', 28.0),
        STRUCT('89394460077ffff', 19.0)
    ],
    8, 'exponential'
);
-- -0.92003263973888194

SELECT carto.MORANS_I_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('89394460c37ffff', 28.0),
        STRUCT('89394460077ffff', 19.0)
    ],
    8, 'exponential'
);
-- -0.92003263973888194

SELECT `carto-un`.carto.MORANS_I_H3(input_data, 3, 'exponential')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

SELECT `carto-un-eu`.carto.MORANS_I_H3(input_data, 3, 'exponential')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

SELECT carto.MORANS_I_H3(input_data, 3, 'exponential')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
);

Additional examples

Computing the spatial autocorrelation of POIs locations in Berlin

MORANS_I_QUADBIN_TABLE

MORANS_I_QUADBIN_TABLE(input, output_table, index_col, value_col, size, decay)

Description

This procedure computes the Moran's I spatial autocorrelation from the input table with Quadbin indexes.

input: STRING the query to the data used to compute the coefficient. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
index_col: STRING name of the column with the Quadbin indexes.
value_col: STRING name of the column with the values for each Quadbin cell.
size: INT64 size of the Quadbin k-ring (distance from the origin). This defines the area around each index cell where the distance decay will be applied. If no neighboring cells are found, the weight of the corresponding index cell is set to zero.
decay: STRING decay function to compute the distance decay. Available functions are: uniform, inverse, inverse_square and exponential.

The index_col cannot contain NULL values, otherwise a Invalid input origin error will be returned.

Output

The results are stored in the table named <output_table>, which contains the following column:

morans_i: FLOAT64 Moran's I spatial autocorrelation.

If all cells have no neighbours, then the procedure will fail.

Example

CALL `carto-un`.carto.MORANS_I_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    5,
    'uniform'
);

CALL `carto-un-eu`.carto.MORANS_I_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    5,
    'uniform'
);

CALL carto.MORANS_I_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    5,
    'uniform'
);

MORANS_I_QUADBIN

MORANS_I_QUADBIN(input, size, decay)

Description

This function computes the Moran's I spatial autocorrelation from the input array of Quadbin indexes.

input: ARRAY<STRUCT<index INT64, value FLOAT64>> input data with the indexes and values of the cells.
size: INT64 size of the Quadbin k-ring (distance from the origin). This defines the area around each index cell where the distance decay will be applied. If no neighboring cells are found, the weight of the corresponding index cell is set to zero.
decay: STRING decay function to compute the distance decay. Available functions are: uniform, inverse, inverse_square and exponential.

The input cannot contain NULL indexes values, otherwise a Invalid input origin error will be returned.

Return type

FLOAT64. If all cells have no neighbours, then the function will fail.

Example

SELECT `carto-un`.carto.MORANS_I_QUADBIN(
    [
        STRUCT(5266443791927869439, 51.0),
        STRUCT(5266443791928131583, 28.0),
        STRUCT(5266443791928918015, 19.0)
    ],
    3, 'exponential'
);
-- -0.29665713826808621

SELECT `carto-un-eu`.carto.MORANS_I_QUADBIN(
    [
        STRUCT(5266443791927869439, 51.0),
        STRUCT(5266443791928131583, 28.0),
        STRUCT(5266443791928918015, 19.0)
    ],
    3, 'exponential'
);
-- -0.29665713826808621

SELECT carto.MORANS_I_QUADBIN(
    [
        STRUCT(5266443791927869439, 51.0),
        STRUCT(5266443791928131583, 28.0),
        STRUCT(5266443791928918015, 19.0)
    ],
    3, 'exponential'
);
-- -0.29665713826808621

SELECT `carto-un`.carto.MORANS_I_QUADBIN(input_data, 3, 'exponential')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
)

SELECT `carto-un-eu`.carto.MORANS_I_QUADBIN(input_data, 3, 'exponential')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
)

SELECT carto.MORANS_I_QUADBIN(input_data, 3, 'exponential')
FROM (
    SELECT ARRAY_AGG(STRUCT(index, value)) AS input_data
    FROM mytable
)

LOCAL_MORANS_I_H3_TABLE

LOCAL_MORANS_I_H3_TABLE(input, output_table, index_col, value_col, size, decay, permutations)

Description

This procedure computes the local Moran's I spatial autocorrelation from the input table with H3 indexes. It outputs the H3 index, local Moran's I spatial autocorrelation value, simulated p value psim, Conditional randomization null - expectation EIc, Conditional randomization null - variance VIc, Total randomization null - expectation EI, Total randomization null - variance VI, and the quad HH=1, LL=2, LH=3, HL=4.

input: STRING the query to the data used to compute the coefficient. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
index_col: STRING name of the column with the H3 indexes.
value_col: STRING name of the column with the values for each H3 cell.
size: INT64 size of the H3 k-ring (distance from the origin). This defines the area around each index cell where the distance decay will be applied. If no neighboring cells are found, the weight of the corresponding index cell is set to zero.
decay: STRING decay function to compute the distance decay. Available functions are: uniform, inverse, inverse_square and exponential.
permutations: INT64 number of permutations for the estimation of p-value.

The index_col cannot contain NULL values, otherwise a Invalid input origin error will be returned.

Output

The results are stored in the table named <output_table>, which contains the following columns:

index: STRING H3 index.
value: FLOAT64 local Moran's I spatial autocorrelation.
psim: FLOAT64 simulated p value.
EIc: FLOAT64 conditional randomization null - expectation.
VIc: FLOAT64 conditional randomization null - variance.
EI: FLOAT64 total randomization null - expectation.
VI: FLOAT64 total randomization null - variance.
quad: INT64 HH=1, LL=2, LH=3, HL=4.

Example

CALL `carto-un`.carto.LOCAL_MORANS_I_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    3,
    'exponential',
    100
);

CALL `carto-un-eu`.carto.LOCAL_MORANS_I_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    3,
    'exponential',
    100
);

CALL carto.LOCAL_MORANS_I_H3_TABLE(
    'myproject.mydataset.h3table',
    'myproject.mydataset.outputtable',
    'h3',
    'value'
    3,
    'exponential',
    100
);

LOCAL_MORANS_I_H3

LOCAL_MORANS_I_H3(input, size, decay, permutations)

Description

This function computes the local Moran's I spatial autocorrelation from the input array of H3 indexes. It outputs the H3 index, local Moran's I spatial autocorrelation value, simulated p value psim, Conditional randomization null - expectation EIc, Conditional randomization null - variance VIc, Total randomization null - expectation EI, Total randomization null - variance VI, and the quad HH=1, LL=2, LH=3, HL=4.

input: ARRAY<STRUCT<index STRING, value FLOAT64>> input data with the indexes and values of the cells.
size: INT64 size of the H3 k-ring (distance from the origin). This defines the area around each index cell where the distance decay will be applied. If no neighboring cells are found, the weight of the corresponding index cell is set to zero.
decay: STRING decay function to compute the distance decay. Available functions are: uniform, inverse, inverse_square and exponential.
permutations: INT64 number of permutations for the estimation of p-value.

The input cannot contain NULL indexes values, otherwise a Invalid input origin error will be returned.

Return type

ARRAY<STRUCT<index STRING, value FLOAT64, psim FLOAT64, EIc FLOAT64, VIc FLOAT64, EI FLOAT64, VI FLOAT64, quad INT64>>

Example

SELECT `carto-un`.carto.LOCAL_MORANS_I_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('8939446033bffff', 28.0),
        STRUCT('8939446032bffff', 19.0)
    ],
    3, 'exponential', 100
);
    "index": "8939446032bffff",
    "value": "-0.342921256629947",
    "psim": "0.0099009900990099011",
    "EIc": "-1.0287637698898404",
    "VIc": "0.0",
    "EI": "-1.0",
    "VI": "-0.64721503525401447",
    "quad": "3"
-- {
--   "index": "8939446032bffff",
--   "value": "-0.342921256629947",
--   "psim": "0.0099009900990099011",
--   "EIc": "-1.0287637698898404",
--   "VIc": "0.0",
--   "EI": "-1.0",
--   "VI": "-0.64721503525401447",
--   "quad": "3"
-- },
-- ...

SELECT `carto-un-eu`.carto.LOCAL_MORANS_I_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('8939446033bffff', 28.0),
        STRUCT('8939446032bffff', 19.0)
    ],
    3, 'exponential', 100
);
    "index": "8939446032bffff",
    "value": "-0.342921256629947",
    "psim": "0.0099009900990099011",
    "EIc": "-1.0287637698898404",
    "VIc": "0.0",
    "EI": "-1.0",
    "VI": "-0.64721503525401447",
    "quad": "3"
-- {
--   "index": "8939446032bffff",
--   "value": "-0.342921256629947",
--   "psim": "0.0099009900990099011",
--   "EIc": "-1.0287637698898404",
--   "VIc": "0.0",
--   "EI": "-1.0",
--   "VI": "-0.64721503525401447",
--   "quad": "3"
-- },
-- ...

SELECT carto.LOCAL_MORANS_I_H3(
    [
        STRUCT('89394460323ffff', 51.0),
        STRUCT('8939446033bffff', 28.0),
        STRUCT('8939446032bffff', 19.0)
    ],
    3, 'exponential', 100
);
    "index": "8939446032bffff",
    "value": "-0.342921256629947",
    "psim": "0.0099009900990099011",
    "EIc": "-1.0287637698898404",
    "VIc": "0.0",
    "EI": "-1.0",
    "VI": "-0.64721503525401447",
    "quad": "3"
-- {
--   "index": "8939446032bffff",
--   "value": "-0.342921256629947",
--   "psim": "0.0099009900990099011",
--   "EIc": "-1.0287637698898404",
--   "VIc": "0.0",
--   "EI": "-1.0",
--   "VI": "-0.64721503525401447",
--   "quad": "3"
-- },
-- ...

SELECT `carto-un`.carto.LOCAL_MORANS_I_H3(
    ARRAY(SELECT AS STRUCT index, value FROM mytable),
    3, 'exponential', 100
);

SELECT `carto-un-eu`.carto.LOCAL_MORANS_I_H3(
    ARRAY(SELECT AS STRUCT index, value FROM mytable),
    3, 'exponential', 100
);

SELECT carto.LOCAL_MORANS_I_H3(
    ARRAY(SELECT AS STRUCT index, value FROM mytable),
    3, 'exponential', 100
);

LOCAL_MORANS_I_QUADBIN_TABLE

LOCAL_MORANS_I_QUADBIN_TABLE(input, output_table, index_col, value_col, size, decay, permutations)

Description

This procedure computes the local Moran's I spatial autocorrelation from the input table with Quadbin indexes. It outputs the Quadbin index, local Moran's I spatial autocorrelation value, simulated p value psim, Conditional randomization null - expectation EIc, Conditional randomization null - variance VIc, Total randomization null - expectation EI, Total randomization null - variance VI, and the quad HH=1, LL=2, LH=3, HL=4.

input: STRING the query to the data used to compute the coefficient. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
index_col: STRING name of the column with the Quadbin indexes.
value_col: STRING name of the column with the values for each Quadbin cell.
size: INT64 size of the Quadbin k-ring (distance from the origin). This defines the area around each index cell where the distance decay will be applied. If no neighboring cells are found, the weight of the corresponding index cell is set to zero.
decay: STRING decay function to compute the distance decay. Available functions are: uniform, inverse, inverse_square and exponential.
permutations: INT64 number of permutations for the estimation of p-value.

The index_col cannot contain NULL values, otherwise a Invalid input origin error will be returned.

Output

The results are stored in the table named <output_table>, which contains the following columns:

index: INT64 quadbin index.
value: FLOAT64 local Moran's I spatial autocorrelation.
psim: FLOAT64 simulated p value.
EIc: FLOAT64 conditional randomization null - expectation.
VIc: FLOAT64 conditional randomization null - variance.
EI: FLOAT64 total randomization null - expectation.
VI: FLOAT64 total randomization null - variance.
quad: INT64 HH=1, LL=2, LH=3, HL=4.

Example

CALL `carto-un`.carto.LOCAL_MORANS_I_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    3,
    'exponential',
    100
);

CALL `carto-un-eu`.carto.LOCAL_MORANS_I_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    3,
    'exponential',
    100
);

CALL carto.LOCAL_MORANS_I_QUADBIN_TABLE(
    'myproject.mydataset.quadbintable',
    'myproject.mydataset.outputtable',
    'quadbin',
    'value'
    3,
    'exponential',
    100
);

LOCAL_MORANS_I_QUADBIN

LOCAL_MORANS_I_QUADBIN(input, size, decay, permutations)

Description

This function computes the local Moran's I spatial autocorrelation from the input array of Quadbin indexes. It outputs the Quadbin index, local Moran's I spatial autocorrelation value, simulated p value psim, Conditional randomization null - expectation EIc, Conditional randomization null - variance VIc, Total randomization null - expectation EI, Total randomization null - variance VI, and the quad HH=1, LL=2, LH=3, HL=4.

input: ARRAY<STRUCT<index INT64, value FLOAT64>> input data with the indexes and values of the cells.
size: INT64 size of the Quadbin k-ring (distance from the origin). This defines the area around each index cell where the distance decay will be applied. If no neighboring cells are found, the weight of the corresponding index cell is set to zero.
decay: STRING decay function to compute the distance decay. Available functions are: uniform, inverse, inverse_square and exponential.
permutations: INT64 number of permutations for the estimation of p-value.

The input cannot contain NULL indexes values, otherwise a Invalid input origin error will be returned.

Return type

ARRAY<STRUCT<index INT64, value FLOAT64, psim FLOAT64, EIc FLOAT64, VIc FLOAT64, EI FLOAT64, VI FLOAT64, quad INT64>>

Example

SELECT `carto-un`.carto.LOCAL_MORANS_I_QUADBIN(
    [
        STRUCT(5266443791927869439, 51.0),
        STRUCT(5266443791928131583, 28.0),
        STRUCT(5266443791928918015, 19.0)
    ],
    3, 'exponential', 100
);
-- {
--   "index": "5266443791928918015",
--   "value": "-0.076228184845253524",
--   "psim": "0.0099009900990099011",
--   "EIc": "-0.70361240532717062",
--   "VIc": "0.29943435718277039",
--   "EI": "-0.68393972058572117",
--   "VI": "0.19089112237884748",
--   "quad": "3"
-- },
-- ...

SELECT `carto-un-eu`.carto.LOCAL_MORANS_I_QUADBIN(
    [
        STRUCT(5266443791927869439, 51.0),
        STRUCT(5266443791928131583, 28.0),
        STRUCT(5266443791928918015, 19.0)
    ],
    3, 'exponential', 100
);
-- {
--   "index": "5266443791928918015",
--   "value": "-0.076228184845253524",
--   "psim": "0.0099009900990099011",
--   "EIc": "-0.70361240532717062",
--   "VIc": "0.29943435718277039",
--   "EI": "-0.68393972058572117",
--   "VI": "0.19089112237884748",
--   "quad": "3"
-- },
-- ...

SELECT carto.LOCAL_MORANS_I_QUADBIN(
    [
        STRUCT(5266443791927869439, 51.0),
        STRUCT(5266443791928131583, 28.0),
        STRUCT(5266443791928918015, 19.0)
    ],
    3, 'exponential', 100
);
-- {
--   "index": "5266443791928918015",
--   "value": "-0.076228184845253524",
--   "psim": "0.0099009900990099011",
--   "EIc": "-0.70361240532717062",
--   "VIc": "0.29943435718277039",
--   "EI": "-0.68393972058572117",
--   "VI": "0.19089112237884748",
--   "quad": "3"
-- },
-- ...

SELECT `carto-un`.carto.LOCAL_MORANS_I_QUADBIN(
    ARRAY(SELECT AS STRUCT index, value FROM mytable),
    3, 'exponential', 100
);

SELECT `carto-un-eu`.carto.LOCAL_MORANS_I_QUADBIN(
    ARRAY(SELECT AS STRUCT index, value FROM mytable),
    3, 'exponential', 100
);

SELECT carto.LOCAL_MORANS_I_QUADBIN(
    ARRAY(SELECT AS STRUCT index, value FROM mytable),
    3, 'exponential', 100
);

VARIOGRAM

VARIOGRAM(input, n_bins, max_distance, model)

Description

This function computes the Variogram from the input array of points and their associated values.

It returns a STRUCT with the parameters of the variogram, the x values, the y values, the predicted y values and the number of values aggregated per bin.

input: ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>> input array with the points and their associated values.
n_bins: INT64 number of bins to compute the semivariance.
max_distance: FLOAT64 maximum distance to compute the semivariance.
model: STRING type of model for fitting the semivariance. It can be either:
- exponential: P0 * (1. - exp(-xi / (P1 / 3.0))) + P2
- spherical: P1 * (1.5 * (xi / P0) - 0.5 * (xi / P0)**3) + P2.

Return type

STRUCT<variogram_params ARRAY<FLOAT64>, x ARRAY<FLOAT64>, y ARRAY<FLOAT64>, yp ARRAY<FLOAT64>, count ARRAY<INT64>>

where:

variogram_params: array containing the parameters [P0, P1, P2] fitted to the model.
x: array with the x values used to fit the model.
y: array with the y values used to fit the model.
yp: array with the y values as predicted by the model.
count: array with the number of elements aggregated in the bin.

Examples

DECLARE sample_points ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>>;

 generate the spatially correlated values
SET sample_points = ARRAY(SELECT AS STRUCT st_geogpoint(lon_sqrt+0.1*RAND(),lat_sqrt+0.1*RAND()) point,
            pow(sin(lon_sqrt)*sin(lat_sqrt),2)+0.1*RAND() value
        FROM
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lat_sqrt
        ORDER BY RAND()
        LIMIT 1000);

 compute parameters of the variogram
SELECT `carto-un`.carto.VARIOGRAM(sample_points, 20, 1.0E5, 'exponential');
-- {
--   variogram_params: [1.8656766501394384, 9890263.713521793, -0.007675798653736552],
--   x: [13433.902872564133, 20772.802451664986, 56973.516169567, 67627.90034684369, 70363.43483710312, 78689.64706974, ...],
--   y: [0.005, 0.125, 3.125, 3.380, 2.0, 2.205, ...],
--   yp: [-0.14889750150153813, 0.49581158712413576, 2.351461086006329, 2.635658071286461, 2.696612846710653, 2.857216896041544, ...],
--   count: [162, 308, 328, 326, 312, 305, ...]
-- }

DECLARE sample_points ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>>;

 generate the spatially correlated values
SET sample_points = ARRAY(SELECT AS STRUCT st_geogpoint(lon_sqrt+0.1*RAND(),lat_sqrt+0.1*RAND()) point,
            pow(sin(lon_sqrt)*sin(lat_sqrt),2)+0.1*RAND() value
        FROM
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lat_sqrt
        ORDER BY RAND()
        LIMIT 1000);

 compute parameters of the variogram
SELECT `carto-un-eu`.carto.VARIOGRAM(sample_points, 20, 1.0E5, 'exponential');
-- {
--   variogram_params: [1.8656766501394384, 9890263.713521793, -0.007675798653736552],
--   x: [13433.902872564133, 20772.802451664986, 56973.516169567, 67627.90034684369, 70363.43483710312, 78689.64706974, ...],
--   y: [0.005, 0.125, 3.125, 3.380, 2.0, 2.205, ...],
--   yp: [-0.14889750150153813, 0.49581158712413576, 2.351461086006329, 2.635658071286461, 2.696612846710653, 2.857216896041544, ...],
--   count: [162, 308, 328, 326, 312, 305, ...]
-- }

DECLARE sample_points ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>>;

 generate the spatially correlated values
SET sample_points = ARRAY(SELECT AS STRUCT st_geogpoint(lon_sqrt+0.1*RAND(),lat_sqrt+0.1*RAND()) point,
            pow(sin(lon_sqrt)*sin(lat_sqrt),2)+0.1*RAND() value
        FROM
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lat_sqrt
        ORDER BY RAND()
        LIMIT 1000);

 compute parameters of the variogram
SELECT carto.VARIOGRAM(sample_points, 20, 1.0E5, 'exponential');
-- {
--   variogram_params: [1.8656766501394384, 9890263.713521793, -0.007675798653736552],
--   x: [13433.902872564133, 20772.802451664986, 56973.516169567, 67627.90034684369, 70363.43483710312, 78689.64706974, ...],
--   y: [0.005, 0.125, 3.125, 3.380, 2.0, 2.205, ...],
--   yp: [-0.14889750150153813, 0.49581158712413576, 2.351461086006329, 2.635658071286461, 2.696612846710653, 2.857216896041544, ...],
--   count: [162, 308, 328, 326, 312, 305, ...]
-- }

ORDINARY_KRIGING_TABLE

ORDINARY_KRIGING_TABLE(input_table, interp_table, target_table, n_bins, max_distance, n_neighbors, model)

Description

This procedure uses Ordinary kriging to compute the interpolated values of a set of points stored in a table, given another set of points with known associated values.

input_table: STRING name of the table with the sample points locations and their values stored in a column named point (type GEOGRAPHY) and value (type FLOAT), respectively. It should be a qualified table name including project and dataset: <project-id>.<dataset-id>.<table-name>.
interp_table: STRING name of the table with the point locations whose values will be interpolated stored in a column named point of type GEOGRAPHY. It should be a qualified table name including project and dataset: <project-id>.<dataset-id>.<table-name>.
target_table: STRING name of the output table where the result of the kriging will be stored. It should be a qualified table name including project and dataset: <project-id>.<dataset-id>.<table-name>. The process will fail if the target table already exists. If NULL, the result will be returned by the procedure and won't be persisted.
n_bins: INT64 number of bins to compute the semivariance.
max_distance: FLOAT64 maximum distance to compute the semivariance.
n_neighbors: INT64 maximum number of neighbors of a point to be taken into account for interpolation.
model: STRING type of model for fitting the semivariance. It can be either:
- exponential: P0 * (1. - exp(-xi / (P1 / 3.0))) + P2
- spherical: P1 * (1.5 * (xi / P0) - 0.5 * (xi / P0)**3) + P2.

Example

CALL `carto-un`.carto.ORDINARY_KRIGING_TABLE(
         'cartobq.docs.nasadem_jp_extract',
         'cartobq.docs.interp_points',
         NULL,
         50,
         1000,
         20,
         'exponential');
-- {"point": "POINT(142.4277 43.51606)", "value": "288.531297133198"},
-- {"point": "POINT(142.4181 43.50518)", "value": "306.62910397500843"},
-- {"point": "POINT(142.4175 43.5045)", "value": "306.9708080004128"},
-- {"point": "POINT(142.4121 43.49838)", "value": "328.37518451985943"},
-- {"point": "POINT(142.4172 43.50416)", "value": "307.1771955935104"},
-- ...

CALL `carto-un-eu`.carto.ORDINARY_KRIGING_TABLE(
         'cartobq.docs.nasadem_jp_extract',
         'cartobq.docs.interp_points',
         NULL,
         50,
         1000,
         20,
         'exponential');
-- {"point": "POINT(142.4277 43.51606)", "value": "288.531297133198"},
-- {"point": "POINT(142.4181 43.50518)", "value": "306.62910397500843"},
-- {"point": "POINT(142.4175 43.5045)", "value": "306.9708080004128"},
-- {"point": "POINT(142.4121 43.49838)", "value": "328.37518451985943"},
-- {"point": "POINT(142.4172 43.50416)", "value": "307.1771955935104"},
-- ...

CALL carto.ORDINARY_KRIGING_TABLE(
         'cartobq.docs.nasadem_jp_extract',
         'cartobq.docs.interp_points',
         NULL,
         50,
         1000,
         20,
         'exponential');
-- {"point": "POINT(142.4277 43.51606)", "value": "288.531297133198"},
-- {"point": "POINT(142.4181 43.50518)", "value": "306.62910397500843"},
-- {"point": "POINT(142.4175 43.5045)", "value": "306.9708080004128"},
-- {"point": "POINT(142.4121 43.49838)", "value": "328.37518451985943"},
-- {"point": "POINT(142.4172 43.50416)", "value": "307.1771955935104"},
-- ...

Additional examples

Interpolating elevation along a road using kriging

ORDINARY_KRIGING

ORDINARY_KRIGING(sample_points, interp_points, max_distance, variogram_params, n_neighbors, model)

Description

This function uses Ordinary kriging to compute the interpolated values of an array of points, given another array of points with known associated values and a variogram. This variogram may be computed with the [#variogram] function.

sample_points: ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>> input array with the sample points and their values.
interp_points: ARRAY<GEOGRAPHY> input array with the points whose values will be interpolated.
max_distance: FLOAT64 maximum distance to compute the semivariance.
variogram_params: ARRAY<FLOAT64> parameters [P0, P1, P2] of the variogram model.
n_neighbors: INT64 maximum number of neighbors of a point to be taken into account for interpolation.
model: STRING type of model for fitting the semivariance. It can be either exponential or spherical and it should be the same type of model as the one used to compute the variogram:
- exponential: P0 * (1. - exp(-xi / (P1 / 3.0))) + P2
- spherical: P1 * (1.5 * (xi / P0) - 0.5 * (xi / P0)**3) + P2.

Return type

ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>>

Examples

Here is a standalone example:

SELECT
  `carto-un`.carto.ORDINARY_KRIGING(
             [STRUCT(st_geogpoint(0.26,1.02) as point, 1.0 as value),
              STRUCT(st_geogpoint(0.91,0.74) as point, 3.1 as value),
              STRUCT(st_geogpoint(-0.59,0.51) as point, 1.5 as value),
              STRUCT(st_geogpoint(0.86,0.92) as point, 3.6 as value),
              STRUCT(st_geogpoint(0.37,1.07) as point, 1.1 as value),
              STRUCT(st_geogpoint(0.69,-0.52) as point, 1.2 as value)],
             [st_geogpoint(0.,0.),
              st_geogpoint(0.,1.)],
             1.0E5,
             [0.1,1E8,0.1],
             20,
             'exponential')
-- {"point": "POINT(0 0)", "value": "1.357680916212768"},
-- {"point": "POINT(0 1)", "value": "1.07161192146499"}

SELECT
  `carto-un-eu`.carto.ORDINARY_KRIGING(
             [STRUCT(st_geogpoint(0.26,1.02) as point, 1.0 as value),
              STRUCT(st_geogpoint(0.91,0.74) as point, 3.1 as value),
              STRUCT(st_geogpoint(-0.59,0.51) as point, 1.5 as value),
              STRUCT(st_geogpoint(0.86,0.92) as point, 3.6 as value),
              STRUCT(st_geogpoint(0.37,1.07) as point, 1.1 as value),
              STRUCT(st_geogpoint(0.69,-0.52) as point, 1.2 as value)],
             [st_geogpoint(0.,0.),
              st_geogpoint(0.,1.)],
             1.0E5,
             [0.1,1E8,0.1],
             20,
             'exponential')
-- {"point": "POINT(0 0)", "value": "1.357680916212768"},
-- {"point": "POINT(0 1)", "value": "1.07161192146499"}

SELECT
  carto.ORDINARY_KRIGING(
             [STRUCT(st_geogpoint(0.26,1.02) as point, 1.0 as value),
              STRUCT(st_geogpoint(0.91,0.74) as point, 3.1 as value),
              STRUCT(st_geogpoint(-0.59,0.51) as point, 1.5 as value),
              STRUCT(st_geogpoint(0.86,0.92) as point, 3.6 as value),
              STRUCT(st_geogpoint(0.37,1.07) as point, 1.1 as value),
              STRUCT(st_geogpoint(0.69,-0.52) as point, 1.2 as value)],
             [st_geogpoint(0.,0.),
              st_geogpoint(0.,1.)],
             1.0E5,
             [0.1,1E8,0.1],
             20,
             'exponential')
-- {"point": "POINT(0 0)", "value": "1.357680916212768"},
-- {"point": "POINT(0 1)", "value": "1.07161192146499"}

Here is an example using the ORDINARY_KRIGING function along with a VARIOGRAM estimation:


DECLARE sample_points ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>>;
DECLARE variogram_output STRUCT<params ARRAY<FLOAT64>, x ARRAY<FLOAT64>, y ARRAY<FLOAT64>, yp ARRAY<FLOAT64>, count ARRAY<INT64>>;
DECLARE interp_points ARRAY<GEOGRAPHY>;

-- Generate the spatially correlated values
SET sample_points = ARRAY(SELECT AS STRUCT st_geogpoint(lon_sqrt+0.1*RAND(),lat_sqrt+0.1*RAND()) point,
            pow(sin(lon_sqrt)*sin(lat_sqrt),2)+0.1*RAND() value
        FROM
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lat_sqrt
        ORDER BY RAND()
        LIMIT 1000);

-- Compute parameters of the variogram
SET variogram_output = `carto-un`.carto.VARIOGRAM(sample_points, 20, 1.0E5, 'spherical');

-- Generate the points to be interpolated
SET interp_points = ARRAY(SELECT st_geogpoint(lon_sqrt,lat_sqrt) point
        FROM
            UNNEST(GENERATE_ARRAY(-5,5,0.25)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-5,5,0.25)) lat_sqrt
            );

-- Calculate interpolated values
SELECT
  point, value
FROM
  UNNEST(`carto-un`.carto.ORDINARY_KRIGING(
         sample_points,
         interp_points,
         1.0E5,
         variogram_output.params,
         20,
         'spherical')) WITH OFFSET pos
ORDER BY pos

-- {"point": POINT(-5 -5), "value": 0.568294714734378},
-- {"point": POINT(-5 -4.75), "value": 0.8303238799265198},
-- {"point": POINT(-5 -4.5), "value": 0.8876712348264676},
-- {"point": POINT(-5 -4.25), "value": 0.7437099678173889},
-- {"point": POINT(-5 -4), "value": 0.5543380644791405},
-- {"point": POINT(-5 -3.75), "value": 0.45182050244159944}
-- ...


DECLARE sample_points ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>>;
DECLARE variogram_output STRUCT<params ARRAY<FLOAT64>, x ARRAY<FLOAT64>, y ARRAY<FLOAT64>, yp ARRAY<FLOAT64>, count ARRAY<INT64>>;
DECLARE interp_points ARRAY<GEOGRAPHY>;

-- Generate the spatially correlated values
SET sample_points = ARRAY(SELECT AS STRUCT st_geogpoint(lon_sqrt+0.1*RAND(),lat_sqrt+0.1*RAND()) point,
            pow(sin(lon_sqrt)*sin(lat_sqrt),2)+0.1*RAND() value
        FROM
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lat_sqrt
        ORDER BY RAND()
        LIMIT 1000);

-- Compute parameters of the variogram
SET variogram_output = `carto-un-eu`.carto.VARIOGRAM(sample_points, 20, 1.0E5, 'spherical');

-- Generate the points to be interpolated
SET interp_points = ARRAY(SELECT st_geogpoint(lon_sqrt,lat_sqrt) point
        FROM
            UNNEST(GENERATE_ARRAY(-5,5,0.25)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-5,5,0.25)) lat_sqrt
            );

-- Calculate interpolated values
SELECT
  point, value
FROM
  UNNEST(`carto-un-eu`.carto.ORDINARY_KRIGING(
         sample_points,
         interp_points,
         1.0E5,
         variogram_output.params,
         20,
         'spherical')) WITH OFFSET pos
ORDER BY pos

-- {"point": POINT(-5 -5), "value": 0.568294714734378},
-- {"point": POINT(-5 -4.75), "value": 0.8303238799265198},
-- {"point": POINT(-5 -4.5), "value": 0.8876712348264676},
-- {"point": POINT(-5 -4.25), "value": 0.7437099678173889},
-- {"point": POINT(-5 -4), "value": 0.5543380644791405},
-- {"point": POINT(-5 -3.75), "value": 0.45182050244159944}
-- ...


DECLARE sample_points ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>>;
DECLARE variogram_output STRUCT<params ARRAY<FLOAT64>, x ARRAY<FLOAT64>, y ARRAY<FLOAT64>, yp ARRAY<FLOAT64>, count ARRAY<INT64>>;
DECLARE interp_points ARRAY<GEOGRAPHY>;

-- Generate the spatially correlated values
SET sample_points = ARRAY(SELECT AS STRUCT st_geogpoint(lon_sqrt+0.1*RAND(),lat_sqrt+0.1*RAND()) point,
            pow(sin(lon_sqrt)*sin(lat_sqrt),2)+0.1*RAND() value
        FROM
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-10,10,0.1)) lat_sqrt
        ORDER BY RAND()
        LIMIT 1000);

-- Compute parameters of the variogram
SET variogram_output = carto.VARIOGRAM(sample_points, 20, 1.0E5, 'spherical');

-- Generate the points to be interpolated
SET interp_points = ARRAY(SELECT st_geogpoint(lon_sqrt,lat_sqrt) point
        FROM
            UNNEST(GENERATE_ARRAY(-5,5,0.25)) lon_sqrt,
            UNNEST(GENERATE_ARRAY(-5,5,0.25)) lat_sqrt
            );

-- Calculate interpolated values
SELECT
  point, value
FROM
  UNNEST(carto.ORDINARY_KRIGING(
         sample_points,
         interp_points,
         1.0E5,
         variogram_output.params,
         20,
         'spherical')) WITH OFFSET pos
ORDER BY pos

-- {"point": POINT(-5 -5), "value": 0.568294714734378},
-- {"point": POINT(-5 -4.75), "value": 0.8303238799265198},
-- {"point": POINT(-5 -4.5), "value": 0.8876712348264676},
-- {"point": POINT(-5 -4.25), "value": 0.7437099678173889},
-- {"point": POINT(-5 -4), "value": 0.5543380644791405},
-- {"point": POINT(-5 -3.75), "value": 0.45182050244159944}
-- ...

IDW

IDW(inputsample, origin, maxdistance, n_neighbors, p)

Description

This function performs Inverse Distance Weighted interpolation. More information on the method can be found here. The method uses the values of the input samples to interpolate the values for the derived locations. The user can select the number of neighbors to be selected for the interpolation, the maximum distance between points and neighbors and the factor p for the weights.

inputsample: ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>> Input array with the sample points and their values.
origin: ARRAY<GEOGRAPHY> Input array with the points whose values will be interpolated.
maxdistance: FLOAT64 Maximum distance between point for interpolation and sampling points.
n_neighbors: INT64 Maximum number of sampling points to be considered for the interpolation.
p: FLOAT64 Power of distance.

Return type

ARRAY<STRUCT<point GEOGRAPHY, value FLOAT64>>

Example

SELECT
      point, value
    FROM
      UNNEST(`carto-un`.carto.IDW(
             ARRAY(SELECT AS STRUCT point, value FROM `cartobq.docs.kriging_sample_points`),
             ARRAY(SELECT point FROM `cartobq.docs.kriging_interp_points`),
             1.0E5,
             20,
             2)) WITH OFFSET pos
    ORDER BY pos
-- TODO

SELECT
      point, value
    FROM
      UNNEST(`carto-un-eu`.carto.IDW(
             ARRAY(SELECT AS STRUCT point, value FROM `cartobq.docs.kriging_sample_points`),
             ARRAY(SELECT point FROM `cartobq.docs.kriging_interp_points`),
             1.0E5,
             20,
             2)) WITH OFFSET pos
    ORDER BY pos
-- TODO

SELECT
      point, value
    FROM
      UNNEST(carto.IDW(
             ARRAY(SELECT AS STRUCT point, value FROM `cartobq.docs.kriging_sample_points`),
             ARRAY(SELECT point FROM `cartobq.docs.kriging_interp_points`),
             1.0E5,
             20,
             2)) WITH OFFSET pos
    ORDER BY pos
-- TODO

SMOOTHING_MRF_H3

SMOOTHING_MRF_H3(input, output, index_column, variable_column, options)

Description

This procedure computes a Markov Random Field (MRF) smoothing for a table containing H3 cell indexes and their associated values.

This implementation is based on the work of Christopher J. Paciorek: "Spatial models for point and areal data using Markov random fields on a fine grid." Electron. J. Statist. 7 946 - 972, 2013. https://doi.org/10.1214/13-EJS791

tip

if your data is in lat/long format, you can still use this procedure by first converting your points to H3 cell indexes by using the H3_FROMLONGLAT function.

input: STRING name of the source table. It should be a fully qualified table name including project and dataset: <project-id>.<dataset-id>.<table-name>.
output: STRING name of the output table. It should be a fully qualified table name including project and dataset: <project-id>.<dataset-id>.<table-name>. The process will fail if the table already exists. If NULL, the result will be returned directly by the procedure and not persisted.
index_column: STRING name of the column containing the cell ids.
variable_column: STRING name of the target variable column.
options: STRING JSON string to overwrite the model's default options. If set to NULL or empty, it will use the default values.
- closing_distance: INT64 distance of closing. It defaults to 0. If strictly positive, the algorithm performs a morphological closing on the cells by the closing_distance, defined in number of cells, before performing the smoothing. No closing is performed otherwise.
- output_closing_cell: BOOL controls whether the cells generated by the closing are added to the output. If defaults to FALSE.
- lambda: FLOAT64 iteration update factor. It defaults to 1.6. For more details, see https://doi.org/10.1214/13-EJS791, page 963.
- iter: INT64 number of iterative queries to perform the smoothing. It defaults to 10. Increasing this parameter might help if the convergence_limit is not reached by the end of the procedure's execution. Tip: if this limit has ben reached, the status of the second-to-last step of the procedure will throw an error.
- intra_iter: INT64 number of iterations per query. It defaults to 50. Reducing this parameter might help if a resource error is reached during the procedure's execution.
- convergence_limit: FLOAT64 threshold condition to stop iterations. If this threshold is not reached, then the procedure will finish its execution after the maximum number of iterations (iter) is reached. It defaults to 10e-5. For more details, see https://doi.org/10.1214/13-EJS791, page 963.

Return type

FLOAT64

Example

CALL `carto-un`.carto.SMOOTHING_MRF_H3( "cartobq.docs.airbnb_berlin_h3_qk",
  NULL,
  'h3_z7',
  'price',
  '{"closing_distance":0, "output_closing_cell":"true", "lambda":1.6, "iter":10, "intra_iter":5, "convergence_limit":10e-5}');
-- {"id": 871f18840ffffff, "beta": 64.56696796809489}
-- {"id": 871f18841ffffff, "beta": 62.61498241759014}
-- {"id": 871f18844ffffff, "beta": 65.47069449331353}

CALL `carto-un-eu`.carto.SMOOTHING_MRF_H3( "cartobq.docs.airbnb_berlin_h3_qk",
  NULL,
  'h3_z7',
  'price',
  '{"closing_distance":0, "output_closing_cell":"true", "lambda":1.6, "iter":10, "intra_iter":5, "convergence_limit":10e-5}');
-- {"id": 871f18840ffffff, "beta": 64.56696796809489}
-- {"id": 871f18841ffffff, "beta": 62.61498241759014}
-- {"id": 871f18844ffffff, "beta": 65.47069449331353}

CALL carto.SMOOTHING_MRF_H3( "cartobq.docs.airbnb_berlin_h3_qk",
  NULL,
  'h3_z7',
  'price',
  '{"closing_distance":0, "output_closing_cell":"true", "lambda":1.6, "iter":10, "intra_iter":5, "convergence_limit":10e-5}');
-- {"id": 871f18840ffffff, "beta": 64.56696796809489}
-- {"id": 871f18841ffffff, "beta": 62.61498241759014}
-- {"id": 871f18844ffffff, "beta": 65.47069449331353}

SMOOTHING_MRF_QUADBIN

SMOOTHING_MRF_QUADBI (input, output, index_column, variable_column, options)

Description

This procedure computes a Markov Random Field (MRF) smoothing for a table containing QUADBIN cell indexes and their associated values.

tip

if your data is in lat/long format, you can still use this procedure by first converting your points to QUADINT cell indexes by using the QUADBIN_FROMLONGLAT function.

input: STRING name of the source table. It should be a fully qualified table name including project and dataset: <project-id>.<dataset-id>.<table-name>.
output: STRING name of the output table. It should be a fully qualified table name including project and dataset: <project-id>.<dataset-id>.<table-name>. The process will fail if the table already exists.
index_column: STRING name of the column containing the cell ids.
variable_column: STRING name of the target variable column.
options: STRING JSON string to overwrite the model's default options. If set to NULL or empty, it will use the default values.
- closing_distance: INT64 distance of closing. It defaults to 0. If strictly positive, the algorithm performs a morphological closing on the cells by the closing_distance, defined in number of cells, before performing the smoothing. No closing is performed otherwise.
- output_closing_cell: BOOL controls whether the cells generated by the closing are added to the output. If defaults to FALSE.
- lambda: FLOAT64 iteration update factor. It defaults to 1.6. For more details, see https://doi.org/10.1214/13-EJS791, page 963.
- iter: INT64 number of iterative queries to perform the smoothing. It defaults to 10. Increasing this parameter might help if the convergence_limit is not reached by the end of the procedure's execution. Tip: if this limit has ben reached, the status of the second-to-last step of the procedure will throw an error.
- intra_iter: INT64 number of iterations per query. It defaults to 50. Reducing this parameter might help if a resource error is reached during the procedure's execution.
- convergence_limit: FLOAT64 threshold condition to stop iterations. If this threshold is not reached, then the procedure will finish its execution after the maximum number of iterations (iter) is reached. It defaults to 10e-5. For more details, see https://doi.org/10.1214/13-EJS791, page 963.

Return type

FLOAT64

Example

CALL `carto-un`.carto.SMOOTHING_MRF_QUADBIN(
  'cartobq.docs.airbnb_berlin_h3_qk_qb',
  'my-project.my-dataset.my-smoothing-table',
  'qb_z11',
  'price',
  '''{
    "closing_distance": 0,
    "output_closing_cell": "true",
    "lambda": 1.6,
    "iter": 10,
    "intra_iter": 5,
    "convergence_limit": 10e-5
  }'''
);
-- The table `my-project.my-dataset.my-smoothing-table` will be created
-- with columns: id, price_smoothed

CALL `carto-un-eu`.carto.SMOOTHING_MRF_QUADBIN(
  'cartobq.docs.airbnb_berlin_h3_qk_qb',
  'my-project.my-dataset.my-smoothing-table',
  'qb_z11',
  'price',
  '''{
    "closing_distance": 0,
    "output_closing_cell": "true",
    "lambda": 1.6,
    "iter": 10,
    "intra_iter": 5,
    "convergence_limit": 10e-5
  }'''
);
-- The table `my-project.my-dataset.my-smoothing-table` will be created
-- with columns: id, price_smoothed

CALL carto.SMOOTHING_MRF_QUADBIN(
  'cartobq.docs.airbnb_berlin_h3_qk_qb',
  'my-project.my-dataset.my-smoothing-table',
  'qb_z11',
  'price',
  '''{
    "closing_distance": 0,
    "output_closing_cell": "true",
    "lambda": 1.6,
    "iter": 10,
    "intra_iter": 5,
    "convergence_limit": 10e-5
  }'''
);
-- The table `my-project.my-dataset.my-smoothing-table` will be created
-- with columns: id, price_smoothed

BUILD_PCAMIX_DATA

BUILD_PCAMIX_DATA(input_query, index_column, cols_num_arr, cols_cat_arr, cols_ord_arr, output_prefix, options)

Description

Prepares the input data for the BUILD_PCAMIX_MODEL procedure.

This procedure is tested against the R package FactoMineR, which adopts the Factorial Analysis of Mixed Data (FAMD) method developed by Pagés (2004). The same method is applied here and generalizes the use of PCA to account for the number of modalities available to each categorical/ordinal variable and on the probabilities of these modalities.

Depending on the variable type, he procedure applies the following transformations to the input data:

For the numerical variables: standard scale the columns to get their z-scores
For the categorical variables:
- One-hot-encode the categorical columns to get their indicator matrix
- Weight each column by the inverse of the square root of its probability, given by the number of ones in each column (Ns) divided by the number of observations (N)
- Center the columns

In this procedure, we have extended this method also to ordinal variables by choosing from different encoding methods and by applying the correspoding weight. The available options include:

Categorical encoding: ordinal variables are hot-encoded and the columns of the resulting indicator matrix are then weighted and centered as in the FAMD method
Numerical encoding: ordinal variables are treated as numerical variables

Input parameters

input_query: STRING the query or the fully qualified name of the table containing the input data which will be used to create the PCA model. It must contain all the individual variables specified in cols_num_arr, cols_cat_arr, and cols_ord_arr.
index_column: STRING the name of the column with the unique geographic identifier.
cols_num_arr: ARRAY<STRING> the array containing the names of the numerical (a.k.a. quantitative) columns. Should be set to NULL if no numerical variables are used.
cols_cat_arr: ARRAY<STRING> the array containing the names of the categorical (a.k.a. qualitative) columns. Should be set to NULL if no categorical variables are used.
cols_ord_arr: ARRAY<STRING> the array containing the names of the ordinal columns. Should be set to NULL if no ordinal variables are used.
output_prefix: STRING destination prefix for the output tables. It must contain the project, dataset and prefix. For example <my-project>.<my-dataset>.<output-prefix>.
options: STRING containing a valid JSON with the different options. Valid options are described in the table below.
Option
Description
ordinal_encoding
STRING the method used to encode ordinal variables. Possible options are CATEGORICAL (DEFAULT): the ordinal variables are treated as categorical variables and are transformed using a one-hot encoding scheme; NUMERICAL: the ordinal variables are treated as numerical variables and used as such, without any additional transformation
new_data_input_query
STRING the query to the data which will be projected in the PCA space to obtain the PC scores. It must contain all the individual variables specified in cols_num_arr, cols_cat_arr, and cols_ord_arr. A qualified table name can be given as well, e.g. 'project-id.dataset-id.table-name'.

Return type

The procedure will output two tables:

Model data table: contains the transformed data for the data that will be used to create the PCA model. The name of the table includes the suffix _model_data, for example ..<output_prefix>_model_data.
New data table: contains the transformed data for the data that will be used to derive the PC scores. The name of the table includes the suffix _new_data, for example ..<output_prefix>_new_data.

Examples

This example shows the call for input data containing a mix of numerical, categorical, ordinal variables:

CALL `carto-un`.carto.BUILD_PCAMIX_DATA(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  ['height', 'distance'], 
  ['winters','shadow','tubers','color'],
  ['soil','preference'],
  '<my-project>.<my-dataset>.<output_prefix>',
  '''{
      "ordinal_encoding":"NUMERICAL"
  }'''
)
-- Table `<my-project>.<my-dataset>.<output_prefix>_model_data` will be created

CALL `carto-un-eu`.carto.BUILD_PCAMIX_DATA(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  ['height', 'distance'], 
  ['winters','shadow','tubers','color'],
  ['soil','preference'],
  '<my-project>.<my-dataset>.<output_prefix>',
  '''{
      "ordinal_encoding":"NUMERICAL"
  }'''
)
-- Table `<my-project>.<my-dataset>.<output_prefix>_model_data` will be created

CALL carto.BUILD_PCAMIX_DATA(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  ['height', 'distance'], 
  ['winters','shadow','tubers','color'],
  ['soil','preference'],
  '<my-project>.<my-dataset>.<output_prefix>',
  '''{
      "ordinal_encoding":"NUMERICAL"
  }'''
)
-- Table `<my-project>.<my-dataset>.<output_prefix>_model_data` will be created

In this example instead, only numerical and categorical variables are included and new data for predicting the principal component scores are specified:

CALL `carto-un`.carto.PCAMIX(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  ['height', 'distance'], 
  ['winters','shadow','tubers','color'],
  ['soil','preference'],
  '<my-project>.<my-dataset>.<output_prefix>',
  '''{
      "ordinal_encoding":"NUMERICAL",
      "new_data_input_query":"SELECT * FROM `cartobq.docs.PCAMIX_new_data_input`"
  }'''
)
-- Tables `<my-project>.<my-dataset>.<output_prefix>_model_data` and `<my-project>.<my-dataset>.<output_prefix>_new_data` will be created

CALL `carto-un-eu`.carto.PCAMIX(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  ['height', 'distance'], 
  ['winters','shadow','tubers','color'],
  ['soil','preference'],
  '<my-project>.<my-dataset>.<output_prefix>',
  '''{
      "ordinal_encoding":"NUMERICAL",
      "new_data_input_query":"SELECT * FROM `cartobq.docs.PCAMIX_new_data_input`"
  }'''
)
-- Tables `<my-project>.<my-dataset>.<output_prefix>_model_data` and `<my-project>.<my-dataset>.<output_prefix>_new_data` will be created

CALL carto.PCAMIX(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  ['height', 'distance'], 
  ['winters','shadow','tubers','color'],
  ['soil','preference'],
  '<my-project>.<my-dataset>.<output_prefix>',
  '''{
      "ordinal_encoding":"NUMERICAL",
      "new_data_input_query":"SELECT * FROM `cartobq.docs.PCAMIX_new_data_input`"
  }'''
)
-- Tables `<my-project>.<my-dataset>.<output_prefix>_model_data` and `<my-project>.<my-dataset>.<output_prefix>_new_data` will be created

BUILD_PCAMIX_MODEL

BUILD_PCAMIX_MODEL(input_query, index_column, output_model, options)

Description

Performs principal component analysis (PCA) of a set of N observations described by a mixture of P categorical, ordinal and numerical variables, also known as Factorial Analysis of Mixed Data. This procedure includes ordinary principal component analysis (PCA), when all the input variables are numerical, and multiple correspondence analysis (MCA), when all the input variables are categorical, as special cases.

Note that when all the P variables are qualitative, the principal component scores are equal to scores of standard MCA times square root of P and the eigenvalues are then equal to the usual eigenvalues of MCA times P. When all the variables are quantitative, the procedure gives exactly the same results as standard PCA.

Technical Note: Factorial Analysis of Mixed Data

Principal Component Analysis (PCA) is primarily suited for continuous data, for which squared differences are well defined, but it also might be applied to discrete variables (although in this case the results might exhibit some artifacts). When dealing with categorical or ordinal data, direct application of PCA is not recommended, even if the data has been hot-encoded, as for example done in the PCA method in Google BigQuery. The issue when applying the PCA method over a table containing the one-hot encoded data is that the component of the variance associated with a categorical/ordinal variable would inherently depend on the number of modalities available to the variable as well as on the probabilities of these modalities, and therefore it would be impossible to equally weight all the input variables when maximizing the variance.

The Factorial Analysis of Mixed Data (FAMD) method can be used to deal with a mix of numerical and categorical data and consists in the folliwing steps:

Transform the input data to standardize the numerical columns and build an indicator matrix for the categorical data scaled to account for the number of modalities available to each category c.f. (BUILD_PCAMIX_DATA)
Apply PCA on the resulting table

Input parameters

input_query: STRING the query to the input data created with the BUILD_PCAMIX_DATA procedure. A qualified table name can be given as well, e.g. 'project-id.dataset-id.table-name'.
index_column: STRING the name of the column with the unique geographic identifier.
output_model: STRING the name for the output model. It should include project and dataset, e.g. 'project-id.dataset-id.table-name'.
options: STRING containing a valid JSON with the different options. Valid options are described in the table below.
Option
Description
NUM_PRINCIPAL_COMPONENTS
INT64 Number of principal components to keep as defined in BigQuery ML CREATE MODEL statement for PCA models
PCA_EXPLAINED_VARIANCE_RATIO
FLOAT64 as defined in BigQuery ML CREATE MODEL statement for PCA models
PCA_SOLVER
STRING as defined in BigQuery ML CREATE MODEL statement for PCA models

Return type

The procedure created a PCA model named <output_model>.

Example

CALL `carto-un`.carto.BUILD_PCAMIX_MODEL(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  '<project-id>.<dataset-id>.<model-name>',
  '''{
      "PCA_EXPLAINED_VARIANCE_RATIO":0.9
  }'''
)
-- Model `<my-project>.<my-dataset>.<model-name>` will be created

CALL `carto-un-eu`.carto.BUILD_PCAMIX_MODEL(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  '<project-id>.<dataset-id>.<model-name>',
  '''{
      "PCA_EXPLAINED_VARIANCE_RATIO":0.9
  }'''
)
-- Model `<my-project>.<my-dataset>.<model-name>` will be created

CALL carto.BUILD_PCAMIX_MODEL(
  '''SELECT * FROM `cartobq.docs.PCAMIX_model_data_input`''',
  'id',
  '<project-id>.<dataset-id>.<model-name>',
  '''{
      "PCA_EXPLAINED_VARIANCE_RATIO":0.9
  }'''
)
-- Model `<my-project>.<my-dataset>.<model-name>` will be created

PREDICT_PCAMIX_SCORES

PREDICT_PCAMIX_MODEL(input_query, index_column, input_model, options)

Description

Given the principal component analysis (PCA) model trained with theBUILD_PCAMIX_MODEL procedure, it returns the principal component scores for the input data, as returned by the BUILD_PCAMIX_DATA procedure.

Input parameters

input_query: STRING the query to the input data created with the BUILD_PCAMIX_DATA procedure which will be used to derive the principal component scores. A qualified table name can be given as well, e.g. 'project-id.dataset-id.table-name'.
index_column: STRING the name of the column with the unique geographic identifier.
input_model: STRING the name for the PCA model trained with the [BUILD_PCAMIX_MODEL])clouds/bigquery/modules/doc/statistics/BUILD_PCAMIX_MODEL.md) procedure. It should include project and dataset, e.g. 'project-id.dataset-id.model-name'.
output_table: STRING the name for the output table. It should include project and dataset, e.g. 'project-id.dataset-id.table-name'.

Return type

The results are stored in the table named <output_table>, which containes

the retained principal component scores, named as principal_component_1, principal_component_2, etc. with the first column being the retained score explaining most of the variance and the last column being the retained score explaining the least of the variance
index_column: the unique geographic identifier. The type of this column depends on the type of index_column in input_query.

Example

CALL `carto-un`.carto.PREDICT_PCAMIX_SCORES(
  '''SELECT * FROM `cartobq.docs.PCAMIX_new_data_input`''',
  'id',
  'cartobq.docs.PCAMIX_model',
  '<project-id>.<dataset-id>.<my-table>'
)
-- Table `<my-project>.<my-dataset>.<my-table>` will be created

CALL `carto-un-eu`.carto.PREDICT_PCAMIX_SCORES(
  '''SELECT * FROM `cartobq.docs.PCAMIX_new_data_input`''',
  'id',
  'cartobq.docs.PCAMIX_model',
  '<project-id>.<dataset-id>.<my-table>'
)
-- Table `<my-project>.<my-dataset>.<my-table>` will be created

CALL carto.PREDICT_PCAMIX_SCORES(
  '''SELECT * FROM `cartobq.docs.PCAMIX_new_data_input`''',
  'id',
  'cartobq.docs.PCAMIX_model',
  '<project-id>.<dataset-id>.<my-table>'
)
-- Table `<my-project>.<my-dataset>.<my-table>` will be created

DETECT_SPATIAL_ANOMALIES

DETECT_SPATIAL_ANOMALIES(input_query, index_column, input_variable_column, output_table, options)

Description

This procedure can be used to detect anomalous spatial regions. It implements the scan statistics framework developed in this R package to detect spatial regions where the variable of interest is higher (or lower) than its baseline value.

Technical Note: The generalized spatial scan statistics framework

The generalized spatial scan statistics framework consists of the following steps:

Choose a set of spatial regions to search over, where each spatial region S consists of a set of spatial locations (e.g. defined using spatial indexes)
Choose models of the data under H0 (the null hypothesis of no cluster of anomalies) and H1(S) (the alternative hypothesis assuming an anomalous cluster in region S). Here we assume that that each location's value is drawn independently from some distribution $Dist(b_{i}, q_{i})$ where $b_{i}$ represents the set of baseline values of that location, and $q_{i}$ represents some underlying relative risk parameter. Second, we make the assumption that the relative risk $q_{i}$ is uniform under the null hypothesis: thus we assume that any space-time variation in the values under the null is accounted for by our baseline parameters and our methods are designed to detect any additional variation not reflected in these baselines.
Choose a baseline. Two typical approaches to obtaining baselines are the following:
- Population-based. In the population-based approach, the observed values are expected to be proportional to the baselines, which typically represent the population corresponding to each spatial location. This population can be either given (e.g. from census data) or inferred (e.g. from sales data), and can be adjusted for any known covariates. Under the simplifying assumption of uniform rates, we wish to test the null hypothesis that the rate is uniform everywhere (all $q_{i,t}$ are equal to some constant $q_{all}$ ) against the set of alternative hypotheses with $q_{i,t} = q_{in}$ inside some region S and $q_{i,t} = q_{out}$ outside S, for some constants $q_{in} > q_{in}$ .
- Expectation-based. In the expectation-based method, the observed values should be equal (and not just proportional as in the population-based approach) to the baseline under the null hypothesis (i.e. each baseline represents the expected value of the input variable for a given location, under the null hypothesis). These expected values are often derived from a regression model used to estimate the expected value given a set of covariates.
Derive a score function F(S) based on H1(S) and H0:

$F(S)=\frac{Pr(Data|H1(S))}{Pr(Data|H0)}$

Find the most interesting regions, i.e. those regions S with the highest values of F(S). Given the above test statistic F(S), this method can be easily applied by choosing a set of regions S, calculating the score function F(S) for each of these regions, and obtaining the highest scoring region(s).
Calculate the statistical significance of each discovered region using a Monte Carlo randomization: each replica is a copy of the original search area where each value is randomly drawn from the null distribution (e.g. for a Poisson and a Gaussian model, we would generate a sample respectively from a Poisson distribution with mean equal to the baseline value and a Gaussian distribution with mean equal to the baseline value and standard deviation equal to the baseline standard deviation). Once we have randomly generated a number of replicas under the null hypothesis, empirical results suggest that the null distribution of the scan statistic is fit well by a Gumbel extreme value distribution G. This implies that we can compute the p-value for each spatial zone following these steps:
- Generate R random permutations of the data
- For each permutation, select the space-time zone associated with the maximum score
- Fit a Gumbel distribution to the maximum scores. Using the methods of moments, the Cumulative Distribution Function (CDF) for the Gumbel distribution of maxima can be calculated as the following: $G(x) = e^{-e^{-\frac{x-\mu}{\beta}}}$ where $\bar{x} = \mu + 0.5772 \, \beta$ and $\sigma = \dfrac{\beta \, \pi}{\sqrt{6}}$
- Compute the p-value as $1-G(x)$ (right-tailed test)

Input

input_query: STRING the query or the fully qualified name of the table containing the data used to detect the spatial anomalies. It must contain the index_column, the input_variable_column, a column named <input_variable_column>_baseline with the values that should be used as a baseline to detect the anomalies and, when the distributional model parameter is set to 'GAUSSIAN' also its variance, its variance, <input_variable_column>_baseline_sigma2. The locations should all have the same timestamps (i.e. no missing locations / time combinations are allowed). No NULL values in any of the input columns are allowed, and no duplicated locations are allowed. Baselines can be broadly defined, depending on the application domain under consideration. For example, the variable of interest might be some counts and the baselines might be the at-risk population of that area. Alternatively, rather than being given the baselines in advance, we might infer these baselines from the data using a statistical model that accounts for any know covariate and can used to estimate the expected values and their variance.
index_column: STRING the name of the column with the unique geographic identifier of the spatial index, either 'H3' or 'QUADBIN'
input_variable_column: STRING the name of the column with the variable for which the spatial anomalies should be detected. When the permutations option is set to a number larger the zero and the values are of the input_variable_column are larger than , scaling of the values of the input_variable_column (and of <input_variable_column>_baseline and <input_variable_column>_baseline_sigma2 columns) is recommended to avoid a floating point error when computing the p-value using the __GUMBEL_PVALUE function
output_table: STRING the fully qualified name of the output table
options: STRING a JSON with the different options. If options is set to NULL then all options are set to default.
Option
Description
Default value
kring_size
ARRAY<INT64> either the minimum and the maximum size of the k-ring used to define the spatial zones (e.g. [0,5]) or an array of specific k-rings (e.g. [0,2,5]). Use the latter to reduce the number of spatial zones and speed up computation
[0, 3]
is_high_mean_anomalies
BOOL a boolean to specify if the analysis is for detecting spatial zones higher (true) or lower (false) than the baseline
true
estimation_method
STRING the estimation method used to detect the spatial anomalies, either 'EXPECTATION' or 'POPULATION' for the expectation- and population-based methods respectively
'EXPECTATION'
distributional_model
STRING the distributional model of the data, either 'POISSON' or 'GAUSSIAN'. The 'POISSON' model should be used only for non-negative data
'GAUSSIAN'
permutations
INT64 the number of permutations used to derive the random replicas to test the anomaly statistical significance
10
max_results
INT64 the maximum number of spatial zones returned. The spatial zones are first ordered in descending order according to the score. When two (or more) space-time zones have the same score, the order is arbitrary
10

Return type

The results are stored in the table named <output_table>, which contains the following columns:

index_scan: STRING the unique identifier of the spatial zone
score: FLOAT64 the score representing the scan statistics
relrisk: FLOAT64 the value of the relative risk of the spatial zone of being anomalous, computed as the ratio of the aggregated observed values to the aggregated baseline values
pgumbel: FLOAT64 the p-value obtained by fitting a Gumbel distribution to the replicate scan statistics
locations: ARRAY<INT64> or ARRAY<STRING> if the index_column is of type QUADBIN and H3 respectively with the distinct locations belonging to the corresponding spatial zone

Examples

This procedure call performs an anomaly detection analysis using the expectation-based method (with a Gaussian distributional model) for spatial zone defined by a k-ring between 1 and 3:

CALL `carto-un`.carto.DETECT_SPATIAL_ANOMALIES(
    '''
    SELECT quadbin, var, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'quadbin',
    'var',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[1,3],
        "distributional_model":"GAUSSIAN",
        "permutations":10,
        "estimation_method":"EXPECTATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.DETECT_SPATIAL_ANOMALIES(
    '''
    SELECT quadbin, var, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'quadbin',
    'var',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[1,3],
        "distributional_model":"GAUSSIAN",
        "permutations":10,
        "estimation_method":"EXPECTATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.DETECT_SPATIAL_ANOMALIES(
    '''
    SELECT quadbin, var, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'quadbin',
    'var',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[1,3],
        "distributional_model":"GAUSSIAN",
        "permutations":10,
        "estimation_method":"EXPECTATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call performs an anomaly detection analysis using the population-based method (with a Poisson distributional model) for specific spatial k-rings:

CALL `carto-un`.carto.DETECT_SPATIAL_ANOMALIES(
    '''
    SELECT h3, counts, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'h3',
    'counts',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[0,1,2],
        "distributional_model":"POISSON",
        "permutations": 100,
        "estimation_method":"POPULATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.DETECT_SPATIAL_ANOMALIES(
    '''
    SELECT h3, counts, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'h3',
    'counts',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[0,1,2],
        "distributional_model":"POISSON",
        "permutations": 100,
        "estimation_method":"POPULATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.DETECT_SPATIAL_ANOMALIES(
    '''
    SELECT h3, counts, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'h3',
    'counts',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[0,1,2],
        "distributional_model":"POISSON",
        "permutations": 100,
        "estimation_method":"POPULATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

DETECT_SPACETIME_ANOMALIES

DETECT_SPACETIME_ANOMALIES(input_query, index_column, date_column, input_variable_column, time_freq, output_table, options)

Description

This procedure can be used to detect anomalous space-time regions. It implements the scan statistics framework developed in this R package to detect space-time regions where the variable of interest is higher (or lower) than its baseline value.

Technical Note: The generalized space-time scan statistics framework

The generalized space-time scan statistics framework consists of the following steps:

Choose a set of spatial regions to search over, where each space-time region S consists of a set of space-time locations (e.g. defined using spatial indexes).
Choose models of the data under H0 (the null hypothesis of no cluster of anomalies) and H1(S) (the alternative hypothesis assuming an anomalous cluster in region S). Here we assume that that each location's value is drawn independently from some distribution $Dist(b_{i,t}, q_{i,t})$ where $b_{i,t}$ represents the set of baseline values of that location, and $q_{i,t}$ represents some underlying relative risk parameter. Second, we make the assumption that the relative risk $q_{i,t}$ is uniform under the null hypothesis: thus we assume that any space-time variation in the values under the null is accounted for by our baseline parameters and our methods are designed to detect any additional variation not reflected in these baselines.
Choose a baseline. Two typical approaches to obtaining baselines are the following:
- Population-based. In the population-based approach, the observed values are expected to be proportional to the baselines, which typically represent the population corresponding to each space-time location. This population can be either given (e.g. from census data) or inferred (e.g. from sales data), and can be adjusted for any known covariates using for exaple a regression model. We wish to test the null hypothesis that the relative risk is uniform everywhere (all $q_{i,t}$ are equal to some constant $q_{all}$ ) against the set of alternative hypotheses with $q_{i,t} = q_{in}$ inside some region S and $q_{i,t} = q_{out}$ outside S, for some constants $q_{in} \stackrel{<}{>} q_{out}$ .
- Expectation-based. In the expectation-based method, the observed values should be equal (and not just proportional as in the population-based approach) to the baseline under the null hypothesis (i.e. each baseline represents the expected value of the input variable for a given location and at a given time stamp, under the null hypothesis). These expected values are often derived from a regression model (e.g. a time series model of the historical data used to forecast the expected value of the current data) or by computing the moving average over a window larger than the expected temporal extent of the anomaly.
Derive a score function F(S) based on H1(S) and H0:

$F(S)=\frac{Pr(Data|H1(S))}{Pr(Data|H0)}$

Find the most interesting regions, i.e. those regions S with the highest values of F(S). Given the above test statistic F(S), this method can be easily applied by choosing a set of regions S, calculating the score function F(S) for each of these regions, and obtaining the highest scoring region(s).
Calculate the statistical significance of each discovered region using a Monte Carlo randomization: each replica is a copy of the original search area where each value is randomly drawn from the null distribution (e.g. for a Poisson and a Gaussian model, we would generate a sample respectively from a Poisson distribution with mean equal to the baseline value and a Gaussian distribution with mean equal to the baseline value and standard deviation equal to the baseline standard deviation). Once we have randomly generated a number of replicas under the null hypothesis, empirical results suggest that the null distribution of the scan statistic is fit well by a Gumbel extreme value distribution G. This implies that we can compute the p-value for each space-time zone following these steps:
- Generate R random permutations of the data
- For each permutation, select the space-time zone associated with the maximum score
- Fit a Gumbel distribution to the maximum scores. Using the methods of moments, the Cumulative Distribution Function (CDF) for the Gumbel distribution of maxima can be calculated as the following: $G(x) = e^{-e^{-\frac{x-\mu}{\beta}}}$ where $\bar{x} = \mu + 0.5772 \, \beta$ and $\sigma = \dfrac{\beta \, \pi}{\sqrt{6}}$
- Compute the p-value as $1-G(x)$ (right-tailed test)

Input

input_query: STRING the query or the fully qualified name of the table containing the data used to detect the space-time anomalies. It must contain the index_column, the date_column, the input_variable_column, a column named <input_variable_column>_baseline with the values that should be used as a baseline to detect the anomalies and, when the distributional model parameter is set to 'GAUSSIAN' also its variance, <input_variable_column>_baseline_sigma2. The locations should all have the same timestamps (i.e. no missing locations / time combinations are allowed). No NULL values in any of the input columns are allowed, and no duplicated location/time pairs are allowed. Baselines can be broadly defined, depending on the application domain under consideration. For example, the variable of interest might be some counts and the baselines might be the at-risk population of that area. Alternatively, rather than being given the baselines in advance, we might infer these baselines from historical data using a time series model that accounts for any know covariate and can used to estimate the expected values and their variance.
index_column: STRING the name of the column with the unique geographic identifier of the spatial index, either 'H3' or 'QUADBIN'
date_column: STRING the name of the column with the timestamp identifier. The type of this column could be any type that can be casted to DATETIME
input_variable_column: STRING the name of the column with the variable for which the space-time anomalies should be detected. When the permutations option is set to a number larger the zero and the values of the input_variable_column are larger than $10^2$ , scaling of the values of the input_variable_column (and of <input_variable_column>_baseline and <input_variable_column>_baseline_sigma2 columns) is recommended to avoid a floating point error when computing the p-value
time_freq: STRING the temporal frequency of the data selected from one of the following: second, minute, hour, day, week, month, quarter, year
output_table: STRING the fully qualified name of the output table
options: STRING a JSON with the different options. If options is set to NULL then all options are set to default.
Option
Description
Default value
kring_size
ARRAY<INT64> either the minimum and the maximum size of the k-ring used to define the spatial zones (e.g. [0,5]) or an array of specific k-rings (e.g. [0,2,5]). Use the latter to reduce the number of spatial zones and speed up computation
[0, 3]
time_bw
ARRAY<INT64> either the minimum and the maximum temporal bandwidth used to define the temporal zones (e.g. [0,10]) or an array of specific temporal bandwidths (e.g. [0,5,10]). Use the latter to reduce the number of temporal zones and speed up computation
[0, L] where L is the maximum length of the time series
is_prospective
BOOL a boolean to specify if the analysis is retrospective or prospective. In a retrospective analysis, the space-time anomalies can happen at any point in time over all the past data (a temporal zone can end at any timestamp); in the prospective analysis instead, only temporal zones that end with the last timestamp are considered and the interest lies in detecting new emerging anomalies
true
is_high_mean_anomalies
BOOL a boolean to specify if the analysis is for detecting space-time zones higher (true) or lower (false) than the baseline
true
estimation_method
STRING the estimation method used to detect the spacetime anomalies, either 'EXPECTATION' or 'POPULATION' for the expectation- and population-based methods respectively. In the population-based method we expect each observed value to be proportional to its baseline under the null hypothesis, while in the expectation-based method, we expect each observed value to be equal to its baseline under the null hypothesis
'EXPECTATION'
distributional_model
STRING the distributional model of the data, either 'POISSON' or 'GAUSSIAN'. The 'POISSON' model should be used only for non-negative data
'GAUSSIAN'
permutations
INT64 the number of permutations used to derive the random replicas to test the anomaly statistical significance
10
max_results
INT64 the maximum number of space-time zones returned. The space-time zones are first ordered in descending order according to the score. When two (or more) space-time zones have the same score, the order is arbitrary
10

Return type

The results are stored in the table named <output_table>, which contains the following columns:

index_scan: STRING the unique identifier of the space-time zone
score: FLOAT64 the score representing the scan statistics
relrisk: FLOAT64 the value of the relative risk of the space-time zone of being anomalous, computed as the ratio of the aggregated observed values to the aggregated baseline values
pgumbel: FLOAT64 the p-value obtained by fitting a Gumbel distribution to the replicate scan statistics
locations: ARRAY<INT64> or ARRAY<STRING> if the index_column is of type QUADBIN and H3 respectively with the distinct locations belonging to the corresponding spatial zone
times: ARRAY<DATETIME> with the distinct timestamps belonging to the corresponding space-time zone

Examples

This procedure call performs a retrospective anomaly detection analysis using the expectation-based method (with a Gaussian distributional model) for every 12-months temporal window and spatial zone defined by a k-ring between 1 and 3:

CALL `carto-un`.carto.DETECT_SPACETIME_ANOMALIES(
    '''
    SELECT quadbin, date, var, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'quadbin',
    'date',
    'var',
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[1,3],
        "time_bw":[0,12],
        "is_prospective": false,
        "distributional_model":"GAUSSIAN",
        "permutations":10,
        "estimation_method":"EXPECTATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.DETECT_SPACETIME_ANOMALIES(
    '''
    SELECT quadbin, date, var, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'quadbin',
    'date',
    'var',
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[1,3],
        "time_bw":[0,12],
        "is_prospective": false,
        "distributional_model":"GAUSSIAN",
        "permutations":10,
        "estimation_method":"EXPECTATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.DETECT_SPACETIME_ANOMALIES(
    '''
    SELECT quadbin, date, var, var_baseline, var_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'quadbin',
    'date',
    'var',
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[1,3],
        "time_bw":[0,12],
        "is_prospective": false,
        "distributional_model":"GAUSSIAN",
        "permutations":10,
        "estimation_method":"EXPECTATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call performs a prospective anomaly detection analysis using the population-based method (with a Poisson distributional model) for specific spatial k-rings and temporal windows:

CALL `carto-un`.carto.DETECT_SPACETIME_ANOMALIES(
    '''
    SELECT h3, date, counts, counts_baseline, counts_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'h3',
    'date',
    'counts',
    'DAY',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[0,1,2],
        "time_bw":[0,10,30],
        "is_prospective": true,
        "distributional_model":"POISSON",
        "permutations": 100,
        "estimation_method":"POPULATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.DETECT_SPACETIME_ANOMALIES(
    '''
    SELECT h3, date, counts, counts_baseline, counts_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'h3',
    'date',
    'counts',
    'DAY',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[0,1,2],
        "time_bw":[0,10,30],
        "is_prospective": true,
        "distributional_model":"POISSON",
        "permutations": 100,
        "estimation_method":"POPULATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.DETECT_SPACETIME_ANOMALIES(
    '''
    SELECT h3, date, counts, counts_baseline, counts_baseline_sigma2
    FROM `<my-project>.<my-dataset>.<my-input-table>`
    ''',
    'h3',
    'date',
    'counts',
    'DAY',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":[0,1,2],
        "time_bw":[0,10,30],
        "is_prospective": true,
        "distributional_model":"POISSON",
        "permutations": 100,
        "estimation_method":"POPULATION"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

AREA_OF_APPLICABILITY

AREA_OF_APPLICABILITY(source_input_query, candidate_input_query, shap_query, index_column, output_prefix, options)

Description

This procedure computes the Area of Applicability (AOA) of a Bigquery ML model. It generates a metric which tells the user where the results from a Machine Learning (ML) model can be trusted when the predictions are extrapolated outside the training space (i.e. where the estimated cross-validation performance holds). Adding a method to compute this metric is particularly useful for non-linear models.

This implementation is based on Meyer, H., & Pebesma, E. (2021). Predicting into unknown space? Estimating the area of applicability of spatial prediction models. Methods in Ecology and Evolution, 12, 1620– 1633.

Given the SHAP values of a trained model, the procedure computes a Dissimilarity Index (DI) for each new data point used for prediction as the multivariate distance between the model covariates for that point and the nearest training data point. To identify those new points that lie in the model AOA, the DI is compared using a threshold obtained as the (outlier-removed) maximum DI of the training data derived via cross-validation: for each training data point the DI is computed as the distance to the nearest training data point that is not in the same (spatial) cross-validation fold with respect to the average of all pairwise distances between all training data. Alternatively, the user can also input a user-defined threshold. To compute the DI, two distance metrics are available:

Euclidean distance: the distance between two data points is computed as the sum over all predictors of the weighted square differences between the standardized value of each predictor variable, where the weight is derived from the model SHAP table. This distance should be used only when all predictor variables are numerical, for which squared differences are well defined.
Gower distance: the distance between two data points is computed as the sum over all predictors of the weighted and normalized absolute differences for numerical (continous and discrete) predictors and the indicator function (0 if equal, 1 otherwise) for categorical/ordinal predictors. This distance can be used for numerical only, categorical only or mixed-type data and is normalized between 0 and 1, with 0 indicating that two points are the same.

The cross-validation folds for the training data can be obtained using a custom index (“CUSTOM_KFOLD”), a random cross-validation strategy (“RANDOM_KFOLD”), or environmental blocking (“ENV_BLOCKING_KFOLD”).

Finally, rows with a NULL value in any of the model predictors are dropped.

Input parameters

source_input_query: STRING the query to the data used to train the model. It must contain all the model predictors columns as well as index_column.
candidate_input_query: STRING query to provide the data over which the domain of applicability is estimated. It must contain all the model predictors columns as well as index_column.
shap_query: STRING the query to the model SHAP table with the feature importance of the model. For example, for the BUILD_REVENUE_MODEL these values are stored in a table with suffix _model_shap. When the model is trained with BigQuery ML the feature importance of the model predictors can also be found on the model Interpretability tag.
index_column: STRING the name of the column with the unique geographic identifier. A column with this name needs to be selected (or created) both in source_input_query and in candidate_input_query.
output_prefix: STRING destination and prefix for the output table. It must contain the project, dataset and prefix: <project>.<dataset>.<prefix>.
options: STRING containing a valid JSON with the different options. Valid options are described in the table below. If options is set to NULL then all options are set to default.
Option
Description
threshold_method
Default: "RANDOM_KFOLD". STRING method used for calculating the threshold to be applied on dissimilarity index of the candidate set in order to identify the area of applicability. Possible options are: "USER_DEFINED_THRESHOLD" uses a user defined threshold to derive the AOA. The threshold is provided by the user-defined threshold value; "CUSTOM_KFOLD" uses a customized k-fold index. The threshold is based on the cross-validation folds stored in the kfold_index_column in the source_input_query data; "RANDOM_KFOLD" uses a random k-fold index. The threshold is based on the cross-validation folds derived from a random k-fold strategy with the number of folds specified by the user in the nfolds parameter; "ENV_BLOCKING_KFOLD" uses a environmental blocking k-fold index. The threshold is based on the cross-validation folds derived from an environmental blocking strategy. This method can only be used when all predictors are numerical, otherwise an error is raised.
threshold
FLOAT64 the user defined threshold when the "USER_DEFINED_THRESHOLD" threshold method is used. The threshold should be defined in the [0,1] interval.
kfold_index_column
STRING name of the cross-validation fold column. If threshold_method is set to "CUSTOM_KFOLD", the user needs to pass this parameter, otherwise an error is raised. If threshold_method is set to "RANDOM_KFOLD" or "ENV_BLOCKING_KFOLD", this parameter is optional.
distance_type
Default: "GOWER". STRING the distance used to compute the dissimilarity index. Possible options are GOWER for the Gower distance and EUCLIDEAN for the Euclidean distance. When working with mixed data types the user can only use the Gower distance, otherwise an error is raised.
outliers_scale_factor
FLOAT64 the scale factor used to define the threshold when threshold_method is set to "CUSTOM_KFOLD", "RANDOM_KFOLD", or "ENV_BLOCKING_KFOLD". Analogue to Tukey’s fences k parameter for outlier detection.
pca_explained_variance_ratio
FLOAT64 the proportion of explained variance retained in the PCA analysis. Only values in the (0,1] range are allowed.
nfolds
INT64 the default number of k-folds when the threshold_method is set to "RANDOM_KFOLD" or "ENV_BLOCKING_KFOLD". Cannot be NULL if threshold_method="RANDOM_KFOLD"; if threshold_method="ENV_BLOCKING_KFOLD", if NULL, nfolds_min and nfolds_max must be specified and the optimal number of folds is computed by deriving the clusters for a number of folds between nfolds_min and nfolds_max and choosing the number of folds (clusters) that minimizes the Calinski-Harabasz Index. If not NULL then nfolds should be at least 1.
nfolds_min
INT64 the minimum number of environmental folds (clusters) if nfolds is set to NULL, otherwise it is ignored.
nfolds_max
INT64 the maximum number of environmental folds (clusters) if nfolds is set to NULL, otherwise it is ignored.
normalize_dissimilarity_index
BOOLEAN if TRUE the dissimilarity factor is normalized between 0 and 1. If threshold_method is set to USER_DEFINED_THRESHOLD this parameter must be set to TRUE.
return_source_dataset
BOOLEAN if TRUE the dissimilarity index for the source model data is also returned. If threshold_method is set to "USER_DEFINED_THRESHOLD" this parameter must be set to FALSE.
The different options for each threshold_method are explained in the table below.
threshold_method
USER_DEFINED_THRESHOLD
CUSTOM_KFOLD
RANDOM_KFOLD
ENV_BLOCKING_KFOLD
Default value
threshold
Mandatory
Ignored
Ignored
Ignored
"RANDOM_KFOLD"
kfold_index_column
Ignored
Mandatory
Optional
Optional
"kfold_index"
distance_type
Optional
Optional
Optional
Optional
"GOWER"
outliers_scale_factor
Ignored
Optional
Optional
Optional
1.5
pca_explained_variance_ratio
Ignored
Ignored
Ignored
Optional
0.9
nfolds
Ignored
Ignored
Mandatory
Optional if nfolds_min and nfolds_max are defined
NULL if OPTIONS IS NOT NULL and 4 otherwise
nfolds_min
Ignored
Ignored
Ignored
Optional if nfolds is defined; mandatory if nfolds_max is defined
NULL
nfolds_max
Ignored
Ignored
Ignored
Optional if nfolds is defined; mandatory if nfolds_min is defined
NULL
normalize_dissimilarity_index
Optional
Optional
Optional
Optional
TRUE
return_source_dataset
Optional
Optional
Optional
Optional
FALSE

Output

The output table with the following columns:

index_column: STRING the unique geographic identifier. The data type of the index_column in source_input_query and in the candidate_input_query is casted to STRING to take into account potential differences between the data type of source_input_query and candidate_input_query.
is_source: BOOLEAN TRUE if a data point is in the source model data and FALSE otherwise (only returned if return_source_dataset is TRUE).
kfold_index_column: STRING the cross-validation fold index for each data point in the source model dataset (only returned if return_source_dataset is TRUE). The name of the column is given by the kfold_index_column parameter in the OPTIONS section.
dissimilarity_index: FLOAT64 the dissimilarity index.
dissimilarity_index_threshold: FLOAT64 the dissimilarity index threshold used to define the Area of Applicability (AOA, data points in the candidate set for which the dissimilarity index is smaller than this threshold belong to the area of applicability).
is_in_area_of_applicability: BOOLEAN TRUE if a data point in the candidate set (is_source = FALSE) is in the AOA and FALSE otherwise. For data points in the source set (is_source = TRUE) this is set to NULL.

Examples

Let's start by setting the OPTIONS to NULL. In this case the threshold is computed from a random k-fold strategy (threshold_method=RANDOM_KFOLD) with nfolds = 4:

CALL `carto-un`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  NULL
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL `carto-un-eu`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  NULL
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  NULL
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

With USER_DEFINED_THRESHOLD, the threshold is provided by the user:

CALL `carto-un`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"USER_DEFINED_THRESHOLD",
    "threshold":0.50
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL `carto-un-eu`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"USER_DEFINED_THRESHOLD",
    "threshold":0.50
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"USER_DEFINED_THRESHOLD",
    "threshold":0.50
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

With CUSTOM_KFOLD, the threshold is based on the cross-validation folds stored in the kfold_index_column in the source_input_query data

CALL `carto-un`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"CUSTOM_KFOLD",
    "kfold_index_column":"kfold_index",
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL `carto-un-eu`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"CUSTOM_KFOLD",
    "kfold_index_column":"kfold_index",
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"CUSTOM_KFOLD",
    "kfold_index_column":"kfold_index",
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

With RANDOM_KFOLD the threshold is based on the cross-validation folds derived from a random k-fold strategy with the number of folds specified by the user in the nfolds parameter

CALL `carto-un`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"RANDOM_KFOLD",
    "nfolds":6,
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL `carto-un-eu`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"RANDOM_KFOLD",
    "nfolds":6,
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"RANDOM_KFOLD",
    "nfolds":6,
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

With ENV_BLOCKING_KFOLD the threshold is based on the cross-validation folds derived from an environmental blocking strategy

CALL `carto-un`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"ENV_BLOCKING_KFOLD",
    "pca_explained_variance_ratio":0.9,
    "nfolds_min":3,
    "nfolds_max":6,
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL `carto-un-eu`.carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"ENV_BLOCKING_KFOLD",
    "pca_explained_variance_ratio":0.9,
    "nfolds_min":3,
    "nfolds_max":6,
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL carto.AREA_OF_APPLICABILITY(
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data` WHERE revenue_avg IS NOT NULL',
  'SELECT * EXCEPT(revenue_avg) FROM `cartobq.docs.aoa_revenue_model_data`',
  'SELECT * FROM `cartobq.docs.aoa_revenue_model_shap`',
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''
  {
    "threshold_method":"ENV_BLOCKING_KFOLD",
    "pca_explained_variance_ratio":0.9,
    "nfolds_min":3,
    "nfolds_max":6,
    "return_source_dataset":true,
    "normalize_dissimilarity_index":true
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

ENV_BLOCKING

ENV_BLOCKING(input_query, predictors, index_column, output_prefix, options)

Description

This procedure derives cross validation (CV) folds based on environmental blocking.

This procedure uses multivariate methods (Principal Component Analysis + K-means clustering) to specify sets of similar conditions based on the input covariates. It should be used to overcome the issue of overfitting due to non-causal predictors: in this case, the spatial structure in the data may be explained by the model through some other non-causal covariate which correlates with the spatial structure. The resulting model predictions may perform fine in a situation where the correlation structure between non-causal and the “true” predictors (i.e. the underlying structures) remains unchanged but they could completely fail when predicting to novel situations (extrapolation).

The method performs a Principal Component Analysis (PCA) on the standardized data and then applies K-means clustering to cluster the data. The cluster number is then used to assign to each data point a corresponding CV fold. If the optimal number of folds is not provided, this is obtained by choosing the number of clusters that minimizes the Calinski-Harabasz Index.

We suggest to limit the use of this procedure to numerical data only. Principal Component Analysis (PCA) and K-means are primarily suited for continuous data, for which squared differences are well defined, but they also might be applied to discrete variables (although in this case the PCA results might exhibit some artifacts and the K-means results will not map back to the data). When dealing with categorical or ordinal data instead, direct application of PCA/K-means is not recommended, even if the data has been one-hot-encoded, as for example done in the PCA and K-means methods in Google BigQuery. The issue when applying the PCA/K-means method over a table containing the one-hot encoded data is that the results would inherently depend on the number of modalities available to each variable as well as on the probabilities of these modalities and therefore it would be impossible to equally weight all the input variables when maximizing the variance (PCA) or the within-cluster sums of squares (K-means).

Finally, rows with a NULL value in any of the model predictors are dropped.

Input parameters

input_query: STRING the input query. It must contain all the model predictors columns.
predictors: ARRAY<STRING> the names of the (numeric) predictors.
index_column: STRING the name of the column with the unique geographic identifier.
output_prefix: STRING destination and prefix for the output table. It must contain the project, dataset and prefix: <project>.<dataset>.<prefix>.
options: STRING containing a valid JSON with the different options. Valid options are described in the table below. If options is set to NULL then all options are set to default.
Option
Description
pca_explained_variance_ratio
FLOAT64 as defined in BigQuery ML CREATE MODEL statement for PCA models (DEFAULT: 0.9).
nfolds
INT64 the default number of folds (clusters). If NULL the optimal number of folds is computed by deriving the clusters for a number of folds between nfolds_min and nfolds_max and choosing the number of folds (clusters) that minimizes the Calinski-Harabasz Index. If not NULL then nfolds should be at least 1.
nfolds_min
INT64 the minimum number of environmental folds (clusters) if nfolds is set to NULL, otherwise it is ignored. If NOT NULL nfolds_min should be at least 1 and nfolds_max should also be specified.
nfolds_max
INT64 the maximum number of environmental folds (clusters) if nfolds is set to NULL, otherwise it is ignored. If NOT NULL nfolds_max should be at always larger than nfolds_min.
kfold_index_column
STRING the name of the cross-validation fold column. If NULL this parameter is set to 'k_fold_index'.

Output

The output table with the following columns:

index_column: STRING the unique geographic identifier. Its type will depend on the type of this column in the input_query.
kfold_index_column: INT64 the cross-validation fold column.

Examples

With nfolds specified:

CALL `carto-un`.carto.ENV_BLOCKING(
  'SELECT * FROM `cartobq.docs.env_blocking_input`',
  ['tavg', 'tmin', 'tmax', 'prec', 'srad', 'vapr', 'wind'],
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''{
    "pca_explained_variance_ratio":0.9,
    "nfolds":4,
    "kfold_index_column":"k_fold_index"
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL `carto-un-eu`.carto.ENV_BLOCKING(
  'SELECT * FROM `cartobq.docs.env_blocking_input`',
  ['tavg', 'tmin', 'tmax', 'prec', 'srad', 'vapr', 'wind'],
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''{
    "pca_explained_variance_ratio":0.9,
    "nfolds":4,
    "kfold_index_column":"k_fold_index"
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL carto.ENV_BLOCKING(
  'SELECT * FROM `cartobq.docs.env_blocking_input`',
  ['tavg', 'tmin', 'tmax', 'prec', 'srad', 'vapr', 'wind'],
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''{
    "pca_explained_variance_ratio":0.9,
    "nfolds":4,
    "kfold_index_column":"k_fold_index"
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

With nfolds_min and nfolds_max specified:

CALL `carto-un`.carto.ENV_BLOCKING(
  'SELECT * FROM `cartobq.docs.env_blocking_input`',
  ['tavg', 'tmin', 'tmax', 'prec', 'srad', 'vapr', 'wind'],
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''{
    "pca_explained_variance_ratio":0.9,
    "nfolds_min":3,
    "nfolds_max":6,
    "kfold_index_column":"k_fold_index"
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL `carto-un-eu`.carto.ENV_BLOCKING(
  'SELECT * FROM `cartobq.docs.env_blocking_input`',
  ['tavg', 'tmin', 'tmax', 'prec', 'srad', 'vapr', 'wind'],
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''{
    "pca_explained_variance_ratio":0.9,
    "nfolds_min":3,
    "nfolds_max":6,
    "kfold_index_column":"k_fold_index"
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

CALL carto.ENV_BLOCKING(
  'SELECT * FROM `cartobq.docs.env_blocking_input`',
  ['tavg', 'tmin', 'tmax', 'prec', 'srad', 'vapr', 'wind'],
  'geoid',
  '<my-project>.<my-dataset>.<output-prefix>',
  '''{
    "pca_explained_variance_ratio":0.9,
    "nfolds_min":3,
    "nfolds_max":6,
    "kfold_index_column":"k_fold_index"
  }
  '''
);
-- Table `<my-project>.<my-dataset>.<output-prefix>` will be created

KRING_SPACETIME_AGGREGATE

KRING_SPACETIME_AGGREGATE(input, index_column, date_column, variables, time_freq, output_table, options)

Description

This procedure can be used to aggregate space-time features within a spatio-temporal neighbourhood to extract new features. It does not resample the input data but allows you to compute statistics for each space-time location based on its neighboring data points. Data must have a temporal dimension and contain grid cell indexes of one of the supported types (H3, Quadbin). For data points to be aggregated with their neighbours, the user must select a k-ring size to define the neighbouring region.

Input

input: STRING the query or the fully qualified name of the table containing the data to be smoothed. It must contain the index_column, the date_column, and all the columns specified in variables.
index_column: STRING the name of the column with the unique geographic identifier of the spatial index, either 'H3' or 'QUADBIN'.
date_column: STRING the name of the column with the timestamp identifier. The type of this column could be any type that can be casted to DATETIME.
variables: ARRAY<STRUCT<variable STRING, aggregation STRING>> the list with the columns and their corresponding aggregation method (sum, avg, max, min, count, mode, perc25,perc50,perc75,perc95). The choice of aggregation function depends on the variable's nature: numerical variables typically use functions like sum, mean, or median (perc50), while categorical variables use modes or counts.
time_freq: STRING the temporal frequency of the data selected from one of the following: second, minute, hour, day, week, month, quarter, year.
output_table: STRING the fully qualified name of the output table.
options: STRING a JSON with the different options. If options is set to NULL then all options are set to default.
Option
Description
Default value
kring_size
INT64 the size of the k-ring used to define the neighbours in the spatial domain; that is, the spatial bandwith. It defines the spatial area (spatial window) around each cell that will be taken into account to aggregate each variable. If set to 0, only the time domain is considered for smoothing the data.
1
time_bw
INT64 the size of the k-ring used to define the neighbours in the time domain; that is, the temporal bandwith. It defines the time range (time window) around each cell that will be taken into account to aggregate each variable. If set to 0, only the spatial domain is considered for smoothing the data.
1
window_alignment
STRING the method used to create the time-based window around each data point that includes its neighboring values. Specifically, for a given point in time, the k-ring encompasses all values within a defined range that can be:
centered: takes all previous and subsequent values within the temporal k-ring, i.e.: [- time_bw, time_bw].
backward: takes all previous values within the temporal k-ring, i.e.: [- time_bw, 0].
forward: takes all subsequent values within the temporal k-ring, i.e.: [0, time_bw].
centered
keep_input_columns
BOOL a boolean to specify whether to preserve the original values in input or to output only the newly smoothened variables.
True

Return type

The results are stored in the table named <output_table>, which contains the columns in input plus the smoothened variables as <variable>_<aggregation>

Examples

This procedure call aggregates variables var1 and var2 within a space-time neighbouring region using a backward time window (i.e. only considering past information). Here we are adding up all neighbouring var1 values, counting all neighbouring var2 cells and counting all neighbouring var2 cells with values larger than 10 (which is enconded in var3).

CALL `carto-un`.carto.KRING_SPACETIME_AGGREGATE(
    "SELECT h3, date, var1, var2, CASE WHEN var2 > 10 THEN 1 ELSE 0 END AS var3 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    [('var1','sum'), ('var2','count'), ('var3','count')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "time_bw":2,
        "window_alignment":"backward"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPACETIME_AGGREGATE(
    "SELECT h3, date, var1, var2, CASE WHEN var2 > 10 THEN 1 ELSE 0 END AS var3 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    [('var1','sum'), ('var2','count'), ('var3','count')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "time_bw":2,
        "window_alignment":"backward"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPACETIME_AGGREGATE(
    "SELECT h3, date, var1, var2, CASE WHEN var2 > 10 THEN 1 ELSE 0 END AS var3 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    [('var1','sum'), ('var2','count'), ('var3','count')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "time_bw":2,
        "window_alignment":"backward"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call aggregates variables var1 and var2 within a space-only neighbouring region.

CALL `carto-un`.carto.KRING_SPACETIME_AGGREGATE(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2,
        "time_bw":0
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPACETIME_AGGREGATE(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2,
        "time_bw":0
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPACETIME_AGGREGATE(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2,
        "time_bw":0
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call aggregates variables var1 and var2 within a time-only neighbouring region.

CALL `carto-un`.carto.KRING_SPACETIME_AGGREGATE(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":0,
        "time_bw":2,
        "window_alignment":"centered"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPACETIME_AGGREGATE(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":0,
        "time_bw":2,
        "window_alignment":"centered"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPACETIME_AGGREGATE(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":0,
        "time_bw":2,
        "window_alignment":"centered"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

KRING_SPACETIME_SMOOTHING

KRING_SPACETIME_SMOOTHING(input, index_column, date_column, value_columns, time_freq, output_table, options)

Description

This procedure can be used to smooth space-time features over space and/or time, aiming to reduce noise or variability in data while maintaining important underlying trends. Data must have a temporal dimension and contain grid cell indexes of one of the supported types (H3, Quadbin). For data points to be averaged with their neighbours, the user must select a k-ring size and a decay function to weight the data of the neighbors cells according to their distance in both time and space (a.k.a. neighboring order).

Input

input: STRING the query or the fully qualified name of the table containing the data to be smoothed. It must contain the index_column, the date_column, and all the columns specified in the value_columns.
index_column: STRING the name of the column with the unique geographic identifier of the spatial index, either 'H3' or 'QUADBIN'.
date_column: STRING the name of the column with the timestamp identifier. The type of this column could be any type that can be casted to DATETIME.
value_columns: ARRAY<STRING> the name of the column(s) with the variable(s) for which the space-time smoothing should be applied. If more than one column is selected, the same smoothing parameters specified in options will be applied to all.
time_freq: STRING the temporal frequency of the data selected from one of the following: second, minute, hour, day, week, month, quarter, year.
output_table: STRING the fully qualified name of the output table.
options: STRING a JSON with the different options. If options is set to NULL then all options are set to default.
Option
Description
Default value
kring_size
INT64 the size of the k-ring used to define the neighbours in the spatial domain; that is, the spatial bandwith. It defines the spatial area (spatial window) around each cell that will be taken into account to average each variable. If set to 0, only the time domain is considered for smoothing the data.
1
time_bw
INT64 the size of the k-ring used to define the neighbours in the time domain; that is, the temporal bandwith. It defines the time range (time window) around each cell that will be taken into account to average each variable. If set to 0, only the spatial domain is considered for smoothing the data.
1
kernel
STRING the kernel function to compute the weights across the spatial kring, that determine how much influence each point within the window has on the smoothed value, based on its distance from the center of the window. All kernel values beyond the bandwidth (kring_size) are set to 0. Available functions are: uniform (all points in each neighborhood are weighted equally), bounded_triangular, bounded_quadratic, bounded_quartic, bounded_gaussian, gaussian, inverse, inverse_square & exponential. Those including the prefix bounded_ are functions whose weights are (close to) zero when the distance equals the bandwidth (i.e. the outermost neighbours).
uniform
kernel_time
STRING the kernel function to compute the weights across the temporal kring, that determine how much influence each point within the window has on the smoothed value, based on its distance from the center of the window. All kernel values beyond the bandwidth (time_bw) are set to 0. Available functions are: uniform (all points in each neighborhood are weighted equally), bounded_triangular, bounded_quadratic, bounded_quartic, bounded_gaussian, gaussian, inverse, inverse_square & exponential. Those including the prefix bounded_ are functions whose weights are (close to) zero when the distance equals the bandwidth (i.e. the outermost neighbours).
uniform
window_alignment
STRING the method used to create the time-based window around each data point that includes its neighboring values. Specifically, for a given point in time, the k-ring encompasses all values within a defined range that can be:
centered: takes all previous and subsequent values within the temporal k-ring, i.e.: [- time_bw, time_bw].
backward: takes all previous values within the temporal k-ring, i.e.: [- time_bw, 0].
forward: takes all subsequent values within the temporal k-ring, i.e.: [0, time_bw].
centered
keep_input_columns
BOOL a boolean to specify whether to preserve the original values in input or to output only the newly smoothened variables.
True

Return type

The results are stored in the table named <output_table>, which contains the columns in input plus the smoothened variables as <value_column>_smooth

Examples

This procedure call smoothens variables var1 and var2 by averaging the values within a space-time neighbouring region, using a backward time window (i.e. only considering past information).

CALL `carto-un`.carto.KRING_SPACETIME_SMOOTHING(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "time_bw":2,
        "kernel":"gaussian",
        "kernel_time":"uniform",
        "window_alignment":"backward"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPACETIME_SMOOTHING(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "time_bw":2,
        "kernel":"gaussian",
        "kernel_time":"uniform",
        "window_alignment":"backward"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPACETIME_SMOOTHING(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "time_bw":2,
        "kernel":"gaussian",
        "kernel_time":"uniform",
        "window_alignment":"backward"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call smoothens variables var1 and var2 by averaging the values within a space-only neighbouring region.

CALL `carto-un`.carto.KRING_SPACETIME_SMOOTHING(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2,
        "time_bw":0,
        "kernel":"gaussian"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPACETIME_SMOOTHING(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2,
        "time_bw":0,
        "kernel":"gaussian"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPACETIME_SMOOTHING(
    "SELECT h3, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2,
        "time_bw":0,
        "kernel":"gaussian"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call smoothens variables var1 and var2 by averaging the values within a time-only neighbouring region.

CALL `carto-un`.carto.KRING_SPACETIME_SMOOTHING(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":0,
        "time_bw":2,
        "kernel_time":"exponential",
        "window_alignment":"centered"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPACETIME_SMOOTHING(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":0,
        "time_bw":2,
        "kernel_time":"exponential",
        "window_alignment":"centered"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPACETIME_SMOOTHING(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":0,
        "time_bw":2,
        "kernel_time":"exponential",
        "window_alignment":"centered"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

KRING_SPATIAL_AGGREGATE

KRING_SPATIAL_AGGREGATE(input, index_column, variables, output_table, options)

Description

This procedure can be used to aggregate spatial features within a neighbourhood to extract new features. It does not resample the input data but allows you to compute statistics for each location based on its neighboring data points. Data must contain grid cell indexes of one of the supported types (H3, Quadbin). For data points to be aggregated with their neighbours, the user must select a k-ring size to define the neighbouring region.

Input

input: STRING the query or the fully qualified name of the table containing the data to be smoothed. It must contain the index_column, and all the columns specified in variables.
index_column: STRING the name of the column with the unique geographic identifier of the spatial index, either 'H3' or 'QUADBIN'.
variables: ARRAY<STRUCT<variable STRING, aggregation STRING>> the list with the columns and their corresponding aggregation method (sum, avg, max, min, count, mode, perc25,perc50,perc75,perc95). The choice of aggregation function depends on the variable's nature: numerical variables typically use functions like sum, mean, or median (perc50), while categorical variables use modes or counts.
output_table: STRING the fully qualified name of the output table.
options: STRING a JSON with the different options. If options is set to NULL then all options are set to default.
Option
Description
Default value
kring_size
INT64 the size of the k-ring used to define the neighbours in the spatial domain; that is, the spatial bandwith. It defines the spatial area (spatial window) around each cell that will be taken into account to aggregate each variable.
1
keep_input_columns
BOOL a boolean to specify whether to preserve the original values in input or to output only the newly smoothened variables.
True

Return type

The results are stored in the table named <output_table>, which contains the columns in input plus the smoothened variables as <variable>_<aggregation>

Examples

This procedure call aggregates variables var1 and var2 within a neighbouring region. Here we are adding up all neighbouring var1 values, counting all neighbouring var2 cells and counting all neighbouring var2 cells with values larger than 10 (which is enconded in var3).

CALL `carto-un`.carto.KRING_SPATIAL_AGGREGATE(
    "SELECT h3, var1, var2, CASE WHEN var2 > 10 THEN 1 ELSE 0 END AS var3 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    [('var1','sum'), ('var2','count'), ('var3','count')],
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPATIAL_AGGREGATE(
    "SELECT h3, var1, var2, CASE WHEN var2 > 10 THEN 1 ELSE 0 END AS var3 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    [('var1','sum'), ('var2','count'), ('var3','count')],
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPATIAL_AGGREGATE(
    "SELECT h3, var1, var2, CASE WHEN var2 > 10 THEN 1 ELSE 0 END AS var3 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    [('var1','sum'), ('var2','count'), ('var3','count')],
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call aggregates variables var1 and var2 within a neighbouring region.

CALL `carto-un`.carto.KRING_SPATIAL_AGGREGATE(
    "SELECT h3, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPATIAL_AGGREGATE(
    "SELECT h3, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPATIAL_AGGREGATE(
    "SELECT h3, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":2
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call aggregates variables var1 and var2 within a neighbouring region.

CALL `carto-un`.carto.KRING_SPATIAL_AGGREGATE(
    "SELECT quadbin, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    '<my-project>.<my-dataset>.<my-output-table>',
    NULL
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPATIAL_AGGREGATE(
    "SELECT quadbin, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    '<my-project>.<my-dataset>.<my-output-table>',
    NULL
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPATIAL_AGGREGATE(
    "SELECT quadbin, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    [('var1','sum'), ('var2','count'), ('var2','mode')],
    '<my-project>.<my-dataset>.<my-output-table>',
    NULL
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

KRING_SPATIAL_SMOOTHING

KRING_SPATIAL_SMOOTHING(input, index_column, value_columns, output_table, options)

Description

This procedure can be used to smooth features over space, aiming to reduce noise or variability in data while maintaining important underlying trends. Data must contain grid cell indexes of one of the supported types (H3, Quadbin). For data points to be averaged with their neighbours, the user must select a k-ring size and a decay function to weight the data of the neighbors cells according to their distance (a.k.a. neighboring order).

Input

input: STRING the query or the fully qualified name of the table containing the data to be smoothed. It must contain the index_column, and all the columns specified in the value_columns.
index_column: STRING the name of the column with the unique geographic identifier of the spatial index, either 'H3' or 'QUADBIN'.
value_columns: ARRAY<STRING> the name of the column(s) with the variable(s) for which the space-time smoothing should be applied. If more than one column is selected, the same smoothing parameters specified in options will be applied to all.
output_table: STRING the fully qualified name of the output table.
options: STRING a JSON with the different options. If options is set to NULL then all options are set to default.
Option
Description
Default value
kring_size
INT64 the size of the k-ring used to define the neighbours in the spatial domain; that is, the spatial bandwith. It defines the spatial area (spatial window) around each cell that will be taken into account to average each variable.
1
kernel
STRING the kernel function to compute the weights across the spatial kring, that determine how much influence each point within the window has on the smoothed value, based on its distance from the center of the window. All kernel values beyond the bandwidth (kring_size) are set to 0. Available functions are: uniform (all points in each neighborhood are weighted equally), bounded_triangular, bounded_quadratic, bounded_quartic, bounded_gaussian, gaussian, inverse, inverse_square & exponential. Those including the prefix bounded_ are functions whose weights are (close to) zero when the distance equals the bandwidth (i.e. the outermost neighbours).
uniform
keep_input_columns
BOOL a boolean to specify whether to preserve the original values in input or to output only the newly smoothened variables.
True

Return type

The results are stored in the table named <output_table>, which contains the columns in input plus the smoothened variables as <value_column>_smooth

Examples

This procedure call smoothens variables var1 and var2 by averaging the values within a spatial neighbouring region.

CALL `carto-un`.carto.KRING_SPATIAL_SMOOTHING(
    "SELECT h3, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    ['var1', 'var2'],
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "kernel":"gaussian"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPATIAL_SMOOTHING(
    "SELECT h3, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    ['var1', 'var2'],
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "kernel":"gaussian"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPATIAL_SMOOTHING(
    "SELECT h3, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'h3',
    ['var1', 'var2'],
    '<my-project>.<my-dataset>.<my-output-table>',
    '''{
        "kring_size":3,
        "kernel":"gaussian"
    }'''
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

This procedure call smoothens variables var1 and var2 by averaging the values within a spatial neighbouring region with defaults.

CALL `carto-un`.carto.KRING_SPATIAL_SMOOTHING(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    NULL
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL `carto-un-eu`.carto.KRING_SPATIAL_SMOOTHING(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    NULL
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

CALL carto.KRING_SPATIAL_SMOOTHING(
    "SELECT quadbin, date, var1, var2 FROM `<my-project>.<my-dataset>.<my-input-table>`",
    'quadbin',
    'date',
    ['var1', 'var2'],
    'MONTH',
    '<my-project>.<my-dataset>.<my-output-table>',
    NULL
)
-- Table `<my-project>.<my-dataset>.<my-output-table>` will be created

SPACETIME_HOTSPOTS_CLASSIFICATION

SPACETIME_HOTSPOTS_CLASSIFICATION (input, output, index_column, date_column, gi_column, p_value_column, options)

Description

This procedure is designed to analyze spatio-temporal data in order to identify and categorize locations based on their hotspot or coldspot status over time. Utilizing z-score values generated by the Space-Time Getis-Ord function, i.e. the h3 spacetime getis ord, and applying either the original Mann Kendall or modified Mann-Kendall trend test on these values, it categorizes each location into specific types of hotspots or coldspots. This categorization is based on patterns of spatial clustering and intensity trends over observed time intervals. The categories can be seen in the following table along with their description.

TIME_SERIES_CLUSTERING

TIME_SERIES_CLUSTERING(
    input,
    output_table,
    partitioning_column,
    ts_column,
    value_column,
    options
)

Description

This procedure provides a way to group together different time series based on several methods that are use-case dependant. The user is able to choose the number of clusters. Please read carefully the method definition below to understand their usage and possible caveats that each method may have.

The data passed to the function requires to be structured using two different columns that will serve as indices:

A unique ID per time series (partitioning_column), which can be a spatial index, a location unique ID (for instance a POI, store, point of sale, antenna, asset, etc.) or any other ID that uniquely identifies each time series.
A timestamp (ts_column), that will identify each of the time steps within each and all series.

All these methods require the series to be aligned and grouped, so that there is one and only one observation per combination of ID and timestamp; the input data cannot have missing values (a series is missing a value in the nth timestep) or multiple values (a series has multiple values in the nth timestep). Since each series require different treatment, the user is in charge of performing this step.

This example below will re-sample a daily time series into a weekly sampling rate and impute all missing values (if any).

WITH
    aligned_idx AS (
        SELECT
            ticket_week,
            shop_id
        FROM
            (
                SELECT DISTINCT
                    DATE_TRUNC(ticket_date, WEEK(MONDAY)) AS ticket_week
                FROM
                    project.dataset.sales_info
            ),
            (
                SELECT DISTINCT
                    shop_id
                FROM
                    project.dataset.sales_info
            )
    )
SELECT
    all.shop_id,
    all.ticket_week
    COALESCE(SUM(info.ticket_amount), 0) AS total_sales
    -- Please note that the sales are an extensive variable; an intensive 
    -- variable would require different aggregation and imputation methods
FROM
    aligned_idx all
LEFT JOIN
    project.dataset.sales_info info
ON
    all.ticket_week = DATE_TRUNC(info.ticket_date, WEEK(MONDAY))
    AND all.shop_id = info.shop_id

WITH
    aligned_idx AS (
        SELECT
            ticket_week,
            shop_id
        FROM
            (
                SELECT DISTINCT
                    DATE_TRUNC(ticket_date, WEEK(MONDAY)) AS ticket_week
                FROM
                    project.dataset.sales_info
            ),
            (
                SELECT DISTINCT
                    shop_id
                FROM
                    project.dataset.sales_info
            )
    )
SELECT
    all.shop_id,
    all.ticket_week
    COALESCE(SUM(info.ticket_amount), 0) AS total_sales
    -- Please note that the sales are an extensive variable; an intensive 
    -- variable would require different aggregation and imputation methods
FROM
    aligned_idx all
LEFT JOIN
    project.dataset.sales_info info
ON
    all.ticket_week = DATE_TRUNC(info.ticket_date, WEEK(MONDAY))
    AND all.shop_id = info.shop_id

WITH
    aligned_idx AS (
        SELECT
            ticket_week,
            shop_id
        FROM
            (
                SELECT DISTINCT
                    DATE_TRUNC(ticket_date, WEEK(MONDAY)) AS ticket_week
                FROM
                    project.dataset.sales_info
            ),
            (
                SELECT DISTINCT
                    shop_id
                FROM
                    project.dataset.sales_info
            )
    )
SELECT
    all.shop_id,
    all.ticket_week
    COALESCE(SUM(info.ticket_amount), 0) AS total_sales
    -- Please note that the sales are an extensive variable; an intensive 
    -- variable would require different aggregation and imputation methods
FROM
    aligned_idx all
LEFT JOIN
    project.dataset.sales_info info
ON
    all.ticket_week = DATE_TRUNC(info.ticket_date, WEEK(MONDAY))
    AND all.shop_id = info.shop_id

The procedure will use any of the methods explained below to cluster the series. There is no single correct method, but different ways to approach different use cases. Please take a look at the official BigQuery guide for working with time series, since probably some of the provided functions or examples can be useful to deal with this kind of preprocessing.

Value characteristic: will group the series based on how similar they are point-wise, that is, sample by sample in each of them. This should return intuitive results: different changes in scale will probably split the series in larger and smaller ones, and changes that make points in similar ranges will contribute to the series being together. Another way of thinking of this method is assuming that will group series together the closer their points are if we plot them on a graph. Note: this function uses the Euclidean Distance internally where the dimensions are the timesteps; therefore it will suffer the Curse of Dimensionality for a very large number of timesteps. To identify this issue, please start the analysis on very large time aggregations (i.e. monthly) and increase the temporal resolution to inspect any changes in this classification.
Profile characteristic: will group the series together based on how similar their dynamics are; that is, how similar their step to step changes are. This method will not take into account their scale but the correlation between series, grouping them together if their changes are similar. For example, two stores will be grouped together if their sales are more substantial on weekends, despite the scale of them may differ orders of magnitude.

Arguments

input: STRING the query to the data used to compute the clustering. A qualified table name can be given as well: <project-id>.<dataset-id>.<table-name>.
output_table: STRING qualified name of the output table: <project-id>.<dataset-id>.<table-name>.
partitioning_column: STRING name of the column with the time series IDs.
ts_column: STRING name of the column with the date.
value_column: STRING name of the column with the value per ID and timestep.
options: JSON containing the advanced options for the procedure:
- method: STRING, one of:
  - value for value characteristic,
  - profile for profile characteristic.
- n_clusters: INT number of clusters to generate in the K-Means.
- model_name: name to use for the K-Means object in BigQuery. Defaults to <output_table>_model.

Output

The results are stored in the table named <output_table>, which contains the following columns:

The given partitioning_column, one entry per unique value in the input;
cluster, a STRING column with the cluster label associated.

Example

CALL `carto-un`.carto.TIME_SERIES_CLUSTERING(
    'myproject.mydataset.input_table',
    'myproject.mydataset.output_table',
    'h3',
    'date',
    'temperature',
    JSON '{ "method": "profile", "n_clusters": 6}'
);

CALL `carto-un-eu`.carto.TIME_SERIES_CLUSTERING(
    'myproject.mydataset.input_table',
    'myproject.mydataset.output_table',
    'h3',
    'date',
    'temperature',
    JSON '{ "method": "profile", "n_clusters": 6}'
);

CALL carto.TIME_SERIES_CLUSTERING(
    'myproject.mydataset.input_table',
    'myproject.mydataset.output_table',
    'h3',
    'date',
    'temperature',
    JSON '{ "method": "profile", "n_clusters": 6}'
);

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 960401.

Previouss2 Nexttelco

Last updated 19 days ago

Was this helpful?