GENERATE_SYNTHETIC_DATA¶

The procedure generates synthetic data from one or more tables, based on data from input tables, and returns a table that contains metrics about the generated data, such as the coefficient of difference (similarity) between the source data and the generated data.

This stored procedure uses the caller’s rights to generate the output table.

Read the requirements for running this procedure. If any requirements are not met, the request will fail before it starts generating data.

Learn more about synthetic data usage.

Syntax¶

SNOWFLAKE.DATA_PRIVACY.GENERATE_SYNTHETIC_DATA(<configuration_object>)

Arguments¶

configuration_object

An OBJECT that specifies the details of the request. You can use an OBJECT constant to specify this object.

The OBJECT value has the following structure:

{
  'datasets': [
    {
      'input_table': '<input_table_name>',
      'output_table' : '<output_table_name>',
      'columns': {
        '<column_name>': {
          <property_name>: <property_value>
        }
        , ...
      }
    }
    , ...
  ],
  'similarity_filter': <boolean>,
  'replace_output_tables': <boolean>,
  'consistency_secret': <session_scoped_reference_string>
}

The OBJECT value contains the following key-value pairs:

datasets

An array specifying the data to generate. Each element in the array is an OBJECT value that defines a single input-output table pair. You can specify a maximum of five table pairs.

A child OBJECT value representing a single input-output table pair has the following properties:

input_table

The fully-qualified name of the input table from which to generate synthetic data. If the table does not exist or cannot be accessed, Snowflake returns an error message. See Using synthetic data in Snowflake for more input table requirements.

If the identifier contains spaces or special characters, the entire string must be enclosed in double quotes. Identifiers enclosed in double quotes are also case-sensitive.

For more information, see Identifier requirements.

output_table

The fully-qualified name of the output table to store the synthetic data generated from input_table. The generated table will have the same permissions and policies as if the user had called CREATE TABLE with default values. If the table already exists and replace_output_tables=TRUE, the existing table will be overwritten.

In addition, the identifier must start with an alphabetic character and cannot contain spaces or special characters unless the entire identifier string is enclosed in double quotes (for example, "My object"). Identifiers enclosed in double quotes are also case-sensitive.

For more information, see Identifier requirements.

columns

(Optional) An OBJECT value specifying additional properties for specific columns. Each field in the OBJECT defines properties for a single column. You do not need to define properties for all columns, or any columns at all. For each field:

The key is the column name. Properties in the value should be applied to this column.

The value is an OBJECT value containing any of the following key-value pairs:

join_key: BOOLEAN, where TRUE indicates that this is a join key column. This cannot be used in a column labeled categorical. Column must be a string, numeric, or Boolean value. Learn about join keys.

Default: FALSE.
categorical: BOOLEAN, used to specify whether the column is a categorical string. Set to TRUE to enable the output to mark the data as non-sensitive, and able to be used in the output. Set to FALSE to redact values from the output. If not specified, will be determined by examining the data. Can be specified only for STRING columns. If set to TRUE, you cannot specify the replace or join_key fields for this column

Default: Inferred based on the column data.

replace: Specifies an output format for STRING values. Can be used only on categorical string columns. The only values that can be used with join_key columns are uuid and email. Cannot be used when categorical is TRUE. If specified, you must provide a value for consistency_secret. The following values are supported:

replace values

Value	Description
`uuid`	A UUID. Example: `88d99a35-c4be-4022-b06a-41fb4629b46d`
`name`	A first and last name in US locale style. Example: `George Washington`
`first_name`	A first name in US locale style. Example: `George`
`last_name`	A last name in US locale style. Example: `Washington`
`address`	An abbreviated address in US locale style. Example: `1600 Pennsylvania Ave`
`full_address`	A detailed street address in US locale style. Example: `1600 Pennsylvania Ave NW, Washington DC 20500`
`email`	An email address. Example: `bdbQ6OPBS5ScOdJx8bVpFw@example.com`
`phone`	A US-style 10-digit phone number in US locale style. Example: `212-555-1234`
`ssn`	A US-style Social Security number. Example: `123-45-6789`

Default:

For join_key columns, uuid
For non-join-key columns, the value will be redacted.

similarity_filter

(Optional) Specifies whether to use a similarity filter when creating the synthetic data. Set this to TRUE to use the built-in privacy filter to remove rows from the target table that are too similar to rows in the input table. If FALSE, each output table will have the same number of rows as its input table; if TRUE, an output table might have fewer rows than its input table. If TRUE, synthetic data generation will fail if you have NULL values in any non-string columns.

Default: FALSE

For more information, see Enhancing privacy.

replace_output_tables

(Optional) Specifies whether to overwrite the output synthetic data table when creating the synthetic data. Set this to TRUE to overwrite the output table.

Default: FALSE

consistency_secret

Session-scoped reference STRING for a symmetric key SECRET. Required if either of the following conditions are met, (otherwise you can omit this field):

If you want consistency for join keys across multiple runs.
If columns.replace or columns.join_key = TRUE are specified on any column, and this procedure is run in an owner’s rights stored procedure.

If you provide a secret, the procedure generates consistent values for STRING join keys across multiple runs that reuse the same consistency secret. If you provide a secret, you must have the READ or OWNERSHIP privilege on this secret.

If you don’t provide a secret, join keys are consistent between tables in the same run, but not across multiple runs. Learn more about consistency.

Default: No consistency

Output¶

Column Name	Data Type	Description
`created_on`	TIMESTAMP	Time the synthetic data was generated.
`table_name`	VARCHAR	Name of the synthetic table.
`table_schema`	VARCHAR	Schema name of the synthetic table.
`table_database`	VARCHAR	Database name of the synthetic table.
`columns`	VARCHAR	A pair of columns in the synthetic table.
`source_table_id`	NUMBER	Internal/system-generated identifier of the input table.
`source_table_name`	VARCHAR	Name of the input table.
`source_table_schema`	VARCHAR	Schema name of the input table.
`source_table_database`	VARCHAR	Database name of the input table.
`source_columns`	VARCHAR	Names of the source columns.
`metric_type`	ENUM	`correlation_coefficient_difference` - Calculated as the absolute value of the correlation coefficient between two non-join columns in the source table and the same two columns in the generated data. Currently, `correlation_coefficient_difference` is the only supported metric. This is the difference between the correlation coefficient of every combination of columns in the input table and the same coefficient in the generated data. Each row represents the correlation coefficient difference between one combination of columns. The column name pair is found in these columns: `columns` and `source_columns`.
`metric_value`	NUMBER	Value of the metric.

Access control requirements¶

To generate synthetic data, you must use a role with each the following grants:

USAGE on the warehouse that you want to use for queries.
SELECT on the input table from which you want to generate synthetic data.
USAGE on the database and schema that contain the input table, and on the database that contains the output table.
CREATE TABLE on the schema that contains the output table.
OWNERSHIP on the output tables. The simplest way to do this is by granting OWNERSHIP to the schema where the output table is generated. (However, if someone has applied a FUTURE GRANT on this schema, table ownership will be silently overridden – that is, GRANT OWNERSHIP ON FUTURE TABLES IN SCHEMA db.my_schema TO ROLE some_role will automatically grant OWNERSHIP to some_role on any new tables created in schema my_schema.)

All users can access the SNOWFLAKE.DATA_PRIVACY.GENERATE_SYNTHETIC_DATA stored procedure. Access is made available using the SNOWFLAKE.CORE_VIEWER database role, which is granted to the PUBLIC role.

Usage notes¶

The JSON key values must be lowercase.
You must accept the Anaconda terms and conditions in your Snowflake account in order to enable this feature.
For additional requirements, see Requirements.
Any timestamps earlier than 1677-09-21 00:12:43.145224193 or later than 2262-04-11 23:47:16.854775807 in the source data are coerced to 1677-09-21 00:12:43.145224193 or 2262-04-11 23:47:16.854775807 respectively when generating synthetic data.

Examples¶

This example generates synthetic data from an input table containing medical information (blood type, gender, age, and ethnicity). The response shows the closeness in data between the source and generated tables. The generated synthetic data table is not shown.

Two columns designated as join keys

CALL SNOWFLAKE.DATA_PRIVACY.GENERATE_SYNTHETIC_DATA({
    'datasets':[
        {
          'input_table': 'syndata_db.sch.faker_source_t',
          'output_table': 'syndata_db.sch.faker_synthetic_t',
          'columns': { 'blood_type': {'join_key': TRUE} , 'ethnicity': {'join_key': TRUE}}
        }
      ]
  });

No columns designated as join keys

CALL SNOWFLAKE.DATA_PRIVACY.GENERATE_SYNTHETIC_DATA({
  'datasets':[
      {
        'input_table': 'syndata_db.sch.faker_source_t',
        'output_table': 'syndata_db.sch.faker_synthetic_t'
      }
    ]
});

Use consistency key to generate consistent values across multiple runs

CREATE OR REPLACE SECRET my_db.public.my_consistency_secret
  TYPE = SYMMETRIC_KEY
  ALGORITHM = GENERIC;

CALL SNOWFLAKE.DATA_PRIVACY.GENERATE_SYNTHETIC_DATA({
  'datasets':[
      {
        'input_table': 'CLINICAL_DB.PUBLIC.BASE_TABLE',
        'output_table': 'my_db.public.test_syndata',
        'columns': { 'patient_id': {'join_key': TRUE, 'replace': 'uuid'}}
      }
    ],
    'consistency_secret': SYSTEM$REFERENCE('SECRET', 'MY_CONSISTENCY_SECRET', 'SESSION', 'READ')::STRING,
    'replace_output_tables': TRUE
});

Output from calling the function

+---------------------------+-------------------+--------------+----------------+------------------------+-------------------+---------------------+-----------------------+------------------------+------------------------------------+----------------+
| CREATED_ON                | TABLE_NAME        | TABLE_SCHEMA | TABLE_DATABASE | COLUMNS                | SOURCE_TABLE_NAME | SOURCE_TABLE_SCHEMA | SOURCE_TABLE_DATABASE | SOURCE_COLUMNS         | METRIC_TYPE                        | METRIC_VALUE   |
+---------------------------+-------------------+--------------+----------------+------------------------+-------------------+---------------------+-----------------------+------------------------+------------------------------------+----------------+
| 2024-07-30 09:53:28.439 Z | faker_synthetic_t | sch          | syndata_db     | "BLOOD_TYPE,GENDER"    | faker_source_t    | sch                 | syndata_db            | "BLOOD_TYPE,GENDER"    | CORRELATION_COEFFICIENT_DIFFERENCE | 0.02430214616  |
| 2024-07-30 09:53:28.439 Z | faker_synthetic_t | sch          | syndata_db     | "BLOOD_TYPE,AGE"       | faker_source_t    | sch                 | syndata_db            | "BLOOD_TYPE,AGE"       | CORRELATION_COEFFICIENT_DIFFERENCE | 0.001919343586 |
| 2024-07-30 09:53:28.439 Z | faker_synthetic_t | sch          | syndata_db     | "BLOOD_TYPE,ETHNICITY" | faker_source_t    | sch                 | syndata_db            | "BLOOD_TYPE,ETHNICITY" | CORRELATION_COEFFICIENT_DIFFERENCE | 0.003720197046 |
| 2024-07-30 09:53:28.439 Z | faker_synthetic_t | sch          | syndata_db     | "GENDER,AGE"           | faker_source_t    | sch                 | syndata_db            | "GENDER,AGE"           | CORRELATION_COEFFICIENT_DIFFERENCE | 0.004348586645 |
| 2024-07-30 09:53:28.439 Z | faker_synthetic_t | sch          | syndata_db     | "GENDER,ETHNICITY"     | faker_source_t    | sch                 | syndata_db            | "GENDER,ETHNICITY"     | CORRELATION_COEFFICIENT_DIFFERENCE | 0.001171535243 |
| 2024-07-30 09:53:28.439 Z | faker_synthetic_t | sch          | syndata_db     | "AGE,ETHNICITY"        | faker_source_t    | sch                 | syndata_db            | "AGE,ETHNICITY"        | CORRELATION_COEFFICIENT_DIFFERENCE | 0.004265938158 |
+---------------------------+-------------------+--------------+----------------+------------------------+-------------------+---------------------+-----------------------+------------------------+------------------------------------+----------------+