The ML.TFDV_DESCRIBE function

Preview

This product or feature is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of the Service Specific Terms. Pre-GA products and features are available "as is" and might have limited support. For more information, see the launch stage descriptions.

Note: To give feedback or request support for this feature, contact [e-mail].

This document describes the ML.TFDV_DESCRIBE function, which you can use to generate fine-grained statistics for the columns in a table. For example, you might want to know statistics for a table of training or serving data statistics that you plan to use with a machine learning (ML) model. Calling this function provides the same behavior as calling the TensorFlow TensorFlow tfdv.generate_statistics_from_csv API. You can use the data output by this function for such purposes as feature preprocessing or model monitoring.

Syntax

ML.TFDV_DESCRIBE(
  { TABLE `project_id.dataset.table` | (query_statement) },
  STRUCT(
    [num_histogram_buckets AS num_histogram_buckets]
    [, num_quantiles_histogram_buckets AS num_quantiles_histogram_buckets]
    [, num_values_histogram_buckets AS num_values_histogram_buckets]
    [, num_rank_histogram_buckets AS num_rank_histogram_buckets])
)

Arguments

ML.TFDV_DESCRIBE takes the following arguments:

• project_id: your project ID.
• dataset: a STRING value that specifies the BigQuery dataset that contains the table.
• table: the name of the input table that contains the training or serving data to calculate statistics for.
• query_statement: a query that generates the training or serving data to calculate statistics for. For the supported SQL syntax of the query_statement clause, see GoogleSQL query syntax.
• num_histogram_buckets: an INT64 value that specifies the number of buckets to use for a histogram with equal-width buckets. Only applies to numerical, ARRAY<numerical>, and ARRAY<STRUCT<INT64, numerical>> columns. The num_histogram_buckets value must be in the range [1, 1,000]. The default value is 10.
• num_quantiles_histogram_buckets: an INT64 value that specifies the number of buckets to use for a quantiles histogram. Only applies to numerical, ARRAY<numerical>, and ARRAY<STRUCT<INT64, numerical>> columns. The num_quantiles_histogram_buckets value must be in the range [1, 1,000]. The default value is 10.
• num_values_histogram_buckets: an INT64 value that specifies the number of buckets to use for a quantiles histogram. Only applies to ARRAY columns. The num_values_histogram_buckets value must be in the range [1, 1,000]. The default value is 10.
• num_rank_histogram_buckets: an INT64 value that specifies the number of buckets to use for a rank histogram. Only applies to categorical and ARRAY<categorical> columns. The num_rank_histogram_buckets value must be in the range [1, 10,000]. The default value is 50.

Output

ML.TFDV_DESCRIBE returns a column named dataset_feature_statistics_list that contains a TensorFlow DatasetFeatureStatisticsList protocol buffer in JSON format.

Example

The following example returns statistics for the penguins public dataset and uses 20 buckets for rank histograms for string values:

SELECT * FROM ML.TFDV_DESCRIBE(
  TABLE `bigquery-public-data.ml_datasets.penguins`,
  STRUCT(20 AS num_rank_histogram_buckets)
);

Limitations

Input data for the ML.TFDV_DESCRIBE function can only contain columns of the following data types:

• Numeric types
• STRING
• BOOL
• BYTE
• DATE
• DATETIME
• TIME
• TIMESTAMP
• ARRAY<STRUCT<INT64, FLOAT64>> (a sparse tensor)
• STRUCT columns that contain any of the following types:
  • Numeric types
  • STRING
  • BOOL
  • BYTE
  • DATE
  • DATETIME
  • TIME
  • TIMESTAMP
• ARRAY columns that contain any of the following types:
  • Numeric types
  • STRING
  • BOOL
  • BYTE
  • DATE
  • DATETIME
  • TIME
  • TIMESTAMP