Quick start#

  1. Create a Data Science notebook session to access jupyterlab interface.

  2. Open a terminal in the notebook session, and then install the fspyspark32_p38_cpu_v3 plugin:

odsc conda install -s fspyspark32_p38_cpu_v3

  1. Download the notebook examples from the example notebook section.

See also

Refer to Notebook Examples which contains more examples for using feature store.

  1. Upload the notebook in the notebook session, and then run the notebook after replacing the required variables.

Feature Store and Data Science Concepts:

Authentication and Policies:

See also

Review the Terraform section for setting up feature store server.

Warning

  1. Feature store doesn’t allow parallel execution of similar logical constructs. Creation is sequential.

  2. If a failure occurs, processing stops, and rollback can’t happen. Retrying the operation isn’t supported.

  3. Define exactly one feature store construct in the YAML file. Creation of multiple feature store constructs in the YAML file causes a failure.

  4. To allow references in the feature store definition, the name of the defined logical constructs in the YAML file must be unique.

  • Python
  • YAML
from ads.feature_store.feature_group_expectation import Expectation, Rule, ExpectationType, ValidationEngineType
from ads.feature_store.feature_store import FeatureStore
from ads.feature_store.input_feature_detail import FeatureDetail, FeatureType
from ads.feature_store.transformation import TransformationMode
import ads

compartment_id = "ocid1.compartment.<unique_id>"
metastore_id = "ocid1.datacatalogmetastore.oc1.iad.<unique_id>"
api_gateway_endpoint = "https://**.{region}.oci.customer-oci.com/20230101"
os.environ["OCI_FS_SERVICE_ENDPOINT"] = api_gateway_endpoint

ads.set_auth(auth="api_key")

# step1: Create the feature store
feature_store_resource = (
    FeatureStore()
    .with_description("<feature_store_description>")
    .with_compartment_id(compartment_id)
    .with_name("<name>")
    .with_offline_config(metastore_id=metastore_id)
)

feature_store = feature_store_resource.create()
entity = feature_store.create_entity(name="product")


# step2: Create the feature store
def transactions_df(dataframe, **kwargs):
    columns = kwargs.get('columns', '*')  # The default is to select all columns if 'columns' isn't provided
    where_clause = kwargs.get('where_clause', '')  # The default is to an empty where clause if 'where_clause' not provided

    sql_query = f"""
                SELECT
                    {columns}
                FROM
                    {table_name}
                {where_clause}
            """
    return sql_query

transformation = feature_store.create_transformation(
    transformation_mode=TransformationMode.SQL,
    source_code_func=transactions_df
)


# step3: Create the expectation
expectation_suite = ExpectationSuite(expectation_suite_name="feature_definition")
expectation_suite.add_expectation(
    ExpectationConfiguration(
        expectation_type="expect_column_values_to_not_be_null",
        kwargs={"column": "date"}
    )
)

input_feature_details = [FeatureDetail("rule_name").with_feature_type(FeatureType.STRING).with_order_number(1)]

# step4: Create the statistics configuration
stats_config = StatisticsConfig().with_is_enabled(False)

# step5: Create the feature group
transformation_args = {"columns": "col1, col2", "where_clause": "col3 > 100"}
feature_group = entity.create_feature_group(
                    primary_keys=["name"],
                    partition_keys=["name"],
                    input_feature_details,
                    expectation_suite=expectation_suite,
                    expectation_type=ExpectationType.LENIENT,
                    statistics_config=stats_config,
                    name="<feature_group_name>",
                    transformation_id=transformation.id,
                    transformation_kwargs=transformation_args
                )

from ads.feature_store.feature_store_registrar import FeatureStoreRegistrar

yaml_string = """
apiVersion: 20230101
kind: featureStore
spec:
  name: *feature_store_name
  offlineConfig:
    metastoreId: *metastore_id

  entity: &entity
    - kind: entity
      spec:
        name: *entity_name


  transformation: &transformation
    - kind: transformation
      spec:
        name: *transformation_name
        transformationMode: *transformation_mode
        sourceCode: *source_code

  featureGroup:
    - kind: featureGroup
      spec:
        name: *feature_group_name
        dataSource: *ds
        description: *feature_group_desc
        transformation: *transformation
        entity: *entity
        primaryKeys:
          *fg_primary_key
        inputFeatureDetails:
          - name: *feature_name
            featureType: *feature_type
            orderNumber: 1

  dataset:
    - kind: dataset
      spec:
        name: *dataset_name
        entity: *entity
        datasetIngestionMode: *ingestion_mode
        description: *dataset_description
        query: *query_statement
"""

feature_registrar = FeatureStoreRegistrar.from_yaml(yaml_string)