Skip to content

SpineGroup#

[source]

SpineGroup#

hsfs.feature_group.SpineGroup(
    storage_connector=None,
    query=None,
    data_format=None,
    path=None,
    options={},
    name=None,
    version=None,
    description=None,
    primary_key=None,
    featurestore_id=None,
    featurestore_name=None,
    created=None,
    creator=None,
    id=None,
    features=None,
    location=None,
    statistics_config=None,
    event_time=None,
    expectation_suite=None,
    online_enabled=False,
    href=None,
    online_topic_name=None,
    spine=True,
    dataframe="spine",
)

Creation#

[source]

get_or_create_spine_group#

FeatureStore.get_or_create_spine_group(
    name, version=None, description="", primary_key=[], event_time=None, features=[], dataframe=None
)

Create a spine group metadata object.

Instead of using a feature group to save a label/prediction target, you can use a spine together with a dataframe containing the labels. A Spine is essentially a metadata object similar to a feature group, however, the data is not materialized in the feature store. It only containes the needed metadata such as the relevant event time column and primary key columns to perform point-in-time correct joins.

Example

# connect to the Feature Store
fs = ...

spine_df = pd.Dataframe()

spine_group = fs.get_or_create_spine_group(
                    name="sales",
                    version=1,
                    description="Physical shop sales features",
                    primary_key=['ss_store_sk'],
                    event_time='sale_date',
                    dataframe=spine_df
                    )

Note that you can inspect the dataframe in the spine group, or replace the dataframe:

spine_group.dataframe.show()

spine_group.dataframe = new_df

The spine can then be used to construct queries, with only one speciality:

Note

Spines can only be used on the left side of a feature join, as this is the base set of entities for which features are to be fetched and the left side of the join determines the event timestamps to compare against.

If you want to use the query for a feature view to be used for online serving, you can only select the label or target feature from the spine. For the online lookup, the label is not required, therefore it is important to only select label from the left feature group, so that we don't need to provide a spine for online serving.

These queries can then be used to create feature views. Since the dataframe contained in the spine is not being materialized, every time you use a feature view created with spine to read data you will have to provide a dataframe with the same structure again.

For example, to generate training data:

X_train, X_test, y_train, y_test = feature_view_spine.train_test_split(0.2, spine=training_data_entities)

Or to get batches of fresh data for batch scoring:

feature_view_spine.get_batch_data(spine=scoring_entities_df).show()

Here you have the chance to pass a different set of entities to generate the training dataset.

Sometimes it might be handy to create a feature view with a regular feature group containing the label, but then at serving time to use a spine in order to fetch features for example only for a small set of primary key values. To do this, you can pass the spine group instead of a dataframe. Just make sure it contains the needed primary key, event time and label column.

feature_view.get_batch_data(spine=spine_group)

Arguments

  • name str: Name of the spine group to create.
  • version Optional[int]: Version of the spine group to retrieve, defaults to None and will create the spine group with incremented version from the last version in the feature store.
  • description Optional[str]: A string describing the contents of the spine group to improve discoverability for Data Scientists, defaults to empty string "".
  • primary_key Optional[List[str]]: A list of feature names to be used as primary key for the spine group. This primary key can be a composite key of multiple features and will be used as joining key, if not specified otherwise. Defaults to empty list [], and the spine group won't have any primary key.
  • event_time Optional[str]: Optionally, provide the name of the feature containing the event time for the features in this spine group. If event_time is set the spine group can be used for point-in-time joins. Defaults to None.
  • features Optional[List[hsfs.feature.Feature]]: Optionally, define the schema of the spine group manually as a list of Feature objects. Defaults to empty list [] and will use the schema information of the DataFrame resulting by executing the provided query against the data source.

Event time data type restriction

The supported data types for the event time column are: timestamp, date and bigint.

  • dataframe Optional[Union[pandas.DataFrame, pyspark.sql.DataFrame, pyspark.RDD, numpy.ndarray, List[list]]]: DataFrame, RDD, Ndarray, list. Spine dataframe with primary key, event time and label column to use for point in time join when fetching features.

Returns

SpineGroup. The spine group metadata object.


Retrieval#

[source]

get_or_create_spine_group#

FeatureStore.get_or_create_spine_group(
    name, version=None, description="", primary_key=[], event_time=None, features=[], dataframe=None
)

Create a spine group metadata object.

Instead of using a feature group to save a label/prediction target, you can use a spine together with a dataframe containing the labels. A Spine is essentially a metadata object similar to a feature group, however, the data is not materialized in the feature store. It only containes the needed metadata such as the relevant event time column and primary key columns to perform point-in-time correct joins.

Example

# connect to the Feature Store
fs = ...

spine_df = pd.Dataframe()

spine_group = fs.get_or_create_spine_group(
                    name="sales",
                    version=1,
                    description="Physical shop sales features",
                    primary_key=['ss_store_sk'],
                    event_time='sale_date',
                    dataframe=spine_df
                    )

Note that you can inspect the dataframe in the spine group, or replace the dataframe:

spine_group.dataframe.show()

spine_group.dataframe = new_df

The spine can then be used to construct queries, with only one speciality:

Note

Spines can only be used on the left side of a feature join, as this is the base set of entities for which features are to be fetched and the left side of the join determines the event timestamps to compare against.

If you want to use the query for a feature view to be used for online serving, you can only select the label or target feature from the spine. For the online lookup, the label is not required, therefore it is important to only select label from the left feature group, so that we don't need to provide a spine for online serving.

These queries can then be used to create feature views. Since the dataframe contained in the spine is not being materialized, every time you use a feature view created with spine to read data you will have to provide a dataframe with the same structure again.

For example, to generate training data:

X_train, X_test, y_train, y_test = feature_view_spine.train_test_split(0.2, spine=training_data_entities)

Or to get batches of fresh data for batch scoring:

feature_view_spine.get_batch_data(spine=scoring_entities_df).show()

Here you have the chance to pass a different set of entities to generate the training dataset.

Sometimes it might be handy to create a feature view with a regular feature group containing the label, but then at serving time to use a spine in order to fetch features for example only for a small set of primary key values. To do this, you can pass the spine group instead of a dataframe. Just make sure it contains the needed primary key, event time and label column.

feature_view.get_batch_data(spine=spine_group)

Arguments

  • name str: Name of the spine group to create.
  • version Optional[int]: Version of the spine group to retrieve, defaults to None and will create the spine group with incremented version from the last version in the feature store.
  • description Optional[str]: A string describing the contents of the spine group to improve discoverability for Data Scientists, defaults to empty string "".
  • primary_key Optional[List[str]]: A list of feature names to be used as primary key for the spine group. This primary key can be a composite key of multiple features and will be used as joining key, if not specified otherwise. Defaults to empty list [], and the spine group won't have any primary key.
  • event_time Optional[str]: Optionally, provide the name of the feature containing the event time for the features in this spine group. If event_time is set the spine group can be used for point-in-time joins. Defaults to None.
  • features Optional[List[hsfs.feature.Feature]]: Optionally, define the schema of the spine group manually as a list of Feature objects. Defaults to empty list [] and will use the schema information of the DataFrame resulting by executing the provided query against the data source.

Event time data type restriction

The supported data types for the event time column are: timestamp, date and bigint.

  • dataframe Optional[Union[pandas.DataFrame, pyspark.sql.DataFrame, pyspark.RDD, numpy.ndarray, List[list]]]: DataFrame, RDD, Ndarray, list. Spine dataframe with primary key, event time and label column to use for point in time join when fetching features.

Returns

SpineGroup. The spine group metadata object.


Properties#

[source]

avro_schema#

Avro schema representation of the feature group.


[source]

dataframe#

Spine dataframe with primary key, event time and label column to use for point in time join when fetching features.


[source]

event_time#

Event time feature in the feature group.


[source]

features#

Schema information.


[source]

primary_key#

List of features building the primary key.


Methods#

[source]

add_tag#

SpineGroup.add_tag(name, value)

Attach a tag to a feature group.

A tag consists of a pair. Tag names are unique identifiers across the whole cluster. The value of a tag can be any valid json - primitives, arrays or json objects.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
fg = fs.get_or_create_feature_group(...)

fg.add_tag(name="example_tag", value="42")

Arguments

  • name str: Name of the tag to be added.
  • value: Value of the tag to be added.

Raises

hsfs.client.exceptions.RestAPIError in case the backend fails to add the tag.


[source]

delete#

SpineGroup.delete()

Drop the entire feature group along with its feature data.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
fg = fs.get_or_create_feature_group(
        name='bitcoin_price',
        version=1
        )

# delete the feature group
fg.delete()

Potentially dangerous operation

This operation drops all metadata associated with this version of the feature group and all the feature data in offline and online storage associated with it.

Raises

hsfs.client.exceptions.RestAPIError.


[source]

delete_tag#

SpineGroup.delete_tag(name)

Delete a tag attached to a feature group.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
fg = fs.get_or_create_feature_group(...)

fg.delete_tag("example_tag")

Arguments

  • name str: Name of the tag to be removed.

Raises

hsfs.client.exceptions.RestAPIError in case the backend fails to delete the tag.


[source]

filter#

SpineGroup.filter(f)

Apply filter to the feature group.

Selects all features and returns the resulting Query with the applied filter.

Example

from hsfs.feature import Feature

# connect to the Feature Store
fs = ...

# get the Feature Group instance
fg = fs.get_or_create_feature_group(...)

fg.filter(Feature("weekly_sales") > 1000)

If you are planning to join the filtered feature group later on with another feature group, make sure to select the filtered feature explicitly from the respective feature group:

Example

fg.filter(fg.feature1 == 1).show(10)

Composite filters require parenthesis:

Example

fg.filter((fg.feature1 == 1) | (fg.feature2 >= 2))

Arguments

  • f Union[hsfs.constructor.filter.Filter, hsfs.constructor.filter.Logic]: Filter object.

Returns

Query. The query object with the applied filter.


[source]

get_complex_features#

SpineGroup.get_complex_features()

Returns the names of all features with a complex data type in this feature group.

Example

complex_dtype_features = fg.get_complex_features()

[source]

get_feature#

SpineGroup.get_feature(name)

Retrieve a Feature object from the schema of the feature group.

There are several ways to access features of a feature group:

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
fg = fs.get_or_create_feature_group(...)

# get Feature instanse
fg.feature1
fg["feature1"]
fg.get_feature("feature1")

Note

Attribute access to features works only for non-reserved names. For example features named id or name will not be accessible via fg.name, instead this will return the name of the feature group itself. Fall back on using the get_feature method.

Arguments:

name: The name of the feature to retrieve

Returns:

Feature: The feature object

Raises

hsfs.client.exceptions.FeatureStoreException.


[source]

get_generated_feature_groups#

SpineGroup.get_generated_feature_groups()

Get the generated feature groups using this feature group, based on explicit provenance. These feature groups can be accessible or inaccessible. Explicit provenance does not track deleted generated feature group links, so deleted will always be empty. For inaccessible feature groups, only a minimal information is returned.

Returns

ProvenanceLinks: Object containing the section of provenance graph requested.

Raises

hsfs.client.exceptions.RestAPIError.


[source]

get_generated_feature_views#

SpineGroup.get_generated_feature_views()

Get the generated feature view using this feature group, based on explicit provenance. These feature views can be accessible or inaccessible. Explicit provenance does not track deleted generated feature view links, so deleted will always be empty. For inaccessible feature views, only a minimal information is returned.

Returns

ProvenanceLinks: Object containing the section of provenance graph requested.

Raises

hsfs.client.exceptions.RestAPIError.


[source]

get_parent_feature_groups#

SpineGroup.get_parent_feature_groups()

Get the parents of this feature group, based on explicit provenance. Parents are feature groups or external feature groups. These feature groups can be accessible, deleted or inaccessible. For deleted and inaccessible feature groups, only a minimal information is returned.

Returns

ProvenanceLinks: Object containing the section of provenance graph requested.

Raises

hsfs.client.exceptions.RestAPIError.


[source]

get_tag#

SpineGroup.get_tag(name)

Get the tags of a feature group.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
fg = fs.get_or_create_feature_group(...)

fg_tag_value = fg.get_tag("example_tag")

Arguments

  • name str: Name of the tag to get.

Returns

tag value

Raises

hsfs.client.exceptions.RestAPIError in case the backend fails to retrieve the tag.


[source]

get_tags#

SpineGroup.get_tags()

Retrieves all tags attached to a feature group.

Returns

Dict[str, obj] of tags.

Raises

hsfs.client.exceptions.RestAPIError in case the backend fails to retrieve the tags.


[source]

json#

SpineGroup.json()

[source]

select#

SpineGroup.select(features=[])

Select a subset of features of the feature group and return a query object.

The query can be used to construct joins of feature groups or create a feature view with a subset of features of the feature group.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
from hsfs.feature import Feature
fg = fs.create_feature_group(
        "fg",
        features=[
                Feature("id", type="string"),
                Feature("ts", type="bigint"),
                Feature("f1", type="date"),
                Feature("f2", type="double")
                ],
        primary_key=["id"],
        event_time="ts")

# construct query
query = fg.select(["id", "f1"])
query.features
# [Feature('id', ...), Feature('f1', ...)]

Arguments

  • features Optional[List[Union[str, hsfs.feature.Feature]]]: A list of Feature objects or feature names as strings to be selected, defaults to [].

Returns

Query: A query object with the selected features of the feature group.


[source]

select_all#

SpineGroup.select_all(include_primary_key=True, include_event_time=True)

Select all features in the feature group and return a query object.

The query can be used to construct joins of feature groups or create a feature view.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instances
fg1 = fs.get_or_create_feature_group(...)
fg2 = fs.get_or_create_feature_group(...)

# construct the query
query = fg1.select_all().join(fg2.select_all())

# show first 5 rows
query.show(5)


# select all features exclude primary key and event time
from hsfs.feature import Feature
fg = fs.create_feature_group(
        "fg",
        features=[
                Feature("id", type="string"),
                Feature("ts", type="bigint"),
                Feature("f1", type="date"),
                Feature("f2", type="double")
                ],
        primary_key=["id"],
        event_time="ts")

query = fg.select_all()
query.features
# [Feature('id', ...), Feature('ts', ...), Feature('f1', ...), Feature('f2', ...)]

query = fg.select_all(include_primary_key=False, include_event_time=False)
query.features
# [Feature('f1', ...), Feature('f2', ...)]

Arguments

  • include_primary_key Optional[bool]: If True, include primary key of the feature group to the feature list. Defaults to True.
  • include_event_time Optional[bool]: If True, include event time of the feature group to the feature list. Defaults to True.

Returns

Query. A query object with all features of the feature group.


[source]

select_except#

SpineGroup.select_except(features=[])

Select all features including primary key and event time feature of the feature group except provided features and return a query object.

The query can be used to construct joins of feature groups or create a feature view with a subset of features of the feature group.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
from hsfs.feature import Feature
fg = fs.create_feature_group(
        "fg",
        features=[
                Feature("id", type="string"),
                Feature("ts", type="bigint"),
                Feature("f1", type="date"),
                Feature("f2", type="double")
                ],
        primary_key=["id"],
        event_time="ts")

# construct query
query = fg.select_except(["ts", "f1"])
query.features
# [Feature('id', ...), Feature('f1', ...)]

Arguments

  • features Optional[List[Union[str, hsfs.feature.Feature]]]: A list of Feature objects or feature names as strings to be excluded from the selection. Defaults to [], selecting all features.

Returns

Query: A query object with the selected features of the feature group.


[source]

to_dict#

SpineGroup.to_dict()

[source]

update_description#

SpineGroup.update_description(description)

Update the description of the feature group.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
fg = fs.get_or_create_feature_group(...)

fg.update_description(description="Much better description.")

Safe update

This method updates the feature group description safely. In case of failure your local metadata object will keep the old description.

Arguments

  • description str: New description string.

Returns

FeatureGroup. The updated feature group object.


[source]

update_feature_description#

SpineGroup.update_feature_description(feature_name, description)

Update the description of a single feature in this feature group.

Example

# connect to the Feature Store
fs = ...

# get the Feature Group instance
fg = fs.get_or_create_feature_group(...)

fg.update_feature_description(feature_name="min_temp",
                              description="Much better feature description.")

Safe update

This method updates the feature description safely. In case of failure your local metadata object will keep the old description.

Arguments

  • feature_name str: Name of the feature to be updated.
  • description str: New description string.

Returns

FeatureGroup. The updated feature group object.


[source]

update_features#

SpineGroup.update_features(features)

Update metadata of features in this feature group.

Currently it's only supported to update the description of a feature.

Unsafe update

Note that if you use an existing Feature object of the schema in the feature group metadata object, this might leave your metadata object in a corrupted state if the update fails.

Arguments

  • features Union[hsfs.feature.Feature, List[hsfs.feature.Feature]]: Feature or list of features. A feature object or list thereof to be updated.

Returns

FeatureGroup. The updated feature group object.