Skip to content

Feature View#

[source]

FeatureView#

hsfs.feature_view.FeatureView(
    name,
    query,
    featurestore_id,
    id=None,
    version=None,
    description="",
    labels=[],
    transformation_functions={},
    featurestore_name=None,
)

Creation#

[source]

create_feature_view#

FeatureStore.create_feature_view(
    name, query, version=None, description="", labels=[], transformation_functions={}
)

Create a feature view metadata object and saved it to hopsworks.

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())

# get the transformation functions
standard_scaler = fs.get_transformation_function(name='standard_scaler')

# construct dictionary of "feature - transformation function" pairs
transformation_functions = {col_name: standard_scaler for col_name in df.columns}

feature_view = fs.create_feature_view(
    name='air_quality_fv',
    version=1,
    transformation_functions=transformation_functions,
    query=query
)

Example

# get feature store instance
fs = ...

# define query object
query = ...

# define dictionary with column names and transformation functions pairs
mapping_transformers = ...

# create feature view
feature_view = fs.create_feature_view(
    name='feature_view_name',
    version=1,
    transformation_functions=mapping_transformers,
    query=query
)

Warning

as_of argument in the Query will be ignored because feature view does not support time travel query.

Arguments

  • name str: Name of the feature view to create.
  • query hsfs.constructor.query.Query: Feature store Query.
  • version Optional[int]: Version of the feature view to create, defaults to None and will create the feature view with incremented version from the last version in the feature store.
  • description Optional[str]: A string describing the contents of the feature view to improve discoverability for Data Scientists, defaults to empty string "".
  • labels Optional[List[str]]: A list of feature names constituting the prediction label/feature of the feature view. When replaying a Query during model inference, the label features can be omitted from the feature vector retrieval. Defaults to [], no label.
  • transformation_functions Optional[Dict[str, hsfs.transformation_function.TransformationFunction]]: A dictionary mapping tansformation functions to to the features they should be applied to before writing out the vector and at inference time. Defaults to {}, no transformations.

Returns:

FeatureView: The feature view metadata object.

Retrieval#

[source]

get_feature_view#

FeatureStore.get_feature_view(name, version=None)

Get a feature view entity from the feature store.

Getting a feature view from the Feature Store means getting its metadata.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(
    name='feature_view_name',
    version=1
)

Arguments

  • name str: Name of the feature view to get.
  • version Optional[int]: Version of the feature view to retrieve, defaults to None and will return the version=1.

Returns

FeatureView: The feature view metadata object.

Raises

  • hsfs.client.exceptions.RestAPIError: If unable to retrieve feature view from the feature store.

[source]

get_feature_views#

FeatureStore.get_feature_views(name)

Get a list of all versions of a feature view entity from the feature store.

Getting a feature view from the Feature Store means getting its metadata.

Example

# get feature store instance
fs = ...

# get a list of all versions of a feature view
feature_view = fs.get_feature_views(
    name='feature_view_name'
)

Arguments

  • name: Name of the feature view to get.

Returns

FeatureView: List of feature view metadata objects.

Raises

  • hsfs.client.exceptions.RestAPIError: If unable to retrieve feature view from the feature store.

Properties#

[source]

description#

Description of the feature view.

[source]

feature_store_name#

Name of the feature store in which the feature group is located.

[source]

featurestore_id#

Feature store id.

[source]

id#

Feature view id.

[source]

labels#

The labels/prediction feature of the feature view.

Can be a composite of multiple features.

[source]

name#

Name of the feature view.

[source]

primary_keys#

Set of primary key names that is used as keys in input dict object for get_serving_vector method.

[source]

query#

Query of the feature view.

[source]

schema#

Feature view schema.

[source]

transformation_functions#

Get transformation functions.

[source]

version#

Version number of the feature view.

Methods#

[source]

add_tag#

FeatureView.add_tag(name, value)

Attach a tag to a feature view.

A tag consists of a name and value 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

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# attach a tag to a feature view
feature_view.add_tag(name="tag_schema", value={"key", "value"})

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]

add_training_dataset_tag#

FeatureView.add_training_dataset_tag(training_dataset_version, name, value)

Attach a tag to a training dataset.

Example

# get feature store instance
fs = ...

# get feature feature view instance
feature_view = fs.get_feature_view(...)

# attach a tag to a training dataset
feature_view.add_training_dataset_tag(
    training_dataset_version=1,
    name="tag_schema",
    value={"key", "value"}
)

Arguments

  • training_dataset_version int: training dataset version
  • 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]

clean#

FeatureView.clean(feature_store_id, feature_view_name, feature_view_version)

Delete the feature view and all associated metadata and training data. This can delete corrupted feature view which cannot be retrieved due to a corrupted query for example.

Example

# delete a feature view and all associated metadata
feature_view.clean(
    feature_store_id=1,
    feature_view_name='feature_view_name',
    feature_view_version=1
)

Potentially dangerous operation

This operation drops all metadata associated with this version of the feature view and related training dataset and materialized data in HopsFS.

Arguments

  • feature_store_id int: int. Id of feature store.
  • feature_view_name str: str. Name of feature view.
  • feature_view_version str: str. Version of feature view.

Raises

hsfs.client.exceptions.RestAPIError.

[source]

create_train_test_split#

FeatureView.create_train_test_split(
    test_size=None,
    train_start="",
    train_end="",
    test_start="",
    test_end="",
    storage_connector=None,
    location="",
    description="",
    extra_filter=None,
    data_format="parquet",
    coalesce=False,
    seed=None,
    statistics_config=None,
    write_options={},
)

Create the metadata for a training dataset and save the corresponding training data into location. The training data is split into train and test set at random or according to time ranges. The training data can be retrieved by calling feature_view.get_train_test_split.

Create random splits

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# create a train-test split dataset
version, job = feature_view.create_train_test_split(
    test_size=0.2,
    description='Description of a dataset',
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format='csv'
)

Create time series splits by specifying date as string

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up dates
train_start = "2022-01-01 00:00:00"
train_end = "2022-06-06 23:59:59"
test_start = "2022-06-07 00:00:00"
test_end = "2022-12-25 23:59:59"

# create a train-test split dataset
version, job = feature_view.create_train_test_split(
    train_start=train_start,
    train_end=train_end,
    test_start=test_start,
    test_end=test_end,
    description='Description of a dataset',
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format='csv'
)

Create time series splits by specifying date as datetime object

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up dates
from datetime import datetime
date_format = "%Y-%m-%d %H:%M:%S"

train_start = datetime.strptime("2022-01-01 00:00:00", date_format)
train_end = datetime.strptime("2022-06-06 23:59:59", date_format)
test_start = datetime.strptime("2022-06-07 00:00:00", date_format)
test_end = datetime.strptime("2022-12-25 23:59:59" , date_format)

# create a train-test split dataset
version, job = feature_view.create_train_test_split(
    train_start=train_start,
    train_end=train_end,
    test_start=test_start,
    test_end=test_end,
    description='Description of a dataset',
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format='csv'
)

Write training dataset to external storage

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get storage connector instance
external_storage_connector = fs.get_storage_connector("storage_connector_name")

# create a train-test split dataset
version, job = feature_view.create_train_test_split(
    train_start=...,
    train_end=...,
    test_start=...,
    test_end=...,
    storage_connector = external_storage_connector,
    description=...,
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format=...
)

Data Formats

The feature store currently supports the following data formats for training datasets:

  1. tfrecord
  2. csv
  3. tsv
  4. parquet
  5. avro
  6. orc

Currently not supported petastorm, hdf5 and npy file formats.

Warning, the following code will fail because category column contains sparse values and training dataset may not have all values available in test split.

import pandas as pd

df = pd.DataFrame({
    'category_col':['category_a','category_b','category_c','category_d'],
    'numeric_col': [40,10,60,40]
})

feature_group = fs.get_or_create_feature_group(
    name='feature_group_name',
    version=1,
    primary_key=['category_col']
)

feature_group.insert(df)

label_encoder = fs.get_transformation_function(name='label_encoder')

feature_view = fs.create_feature_view(
    name='feature_view_name',
    query=feature_group.select_all(),
    transformation_functions={'category_col':label_encoder}
)

feature_view.create_train_test_split(
    test_size=0.5
)
# Output: KeyError: 'category_c'

Arguments

  • test_size Optional[float]: size of test set.
  • train_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the train split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • train_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the train split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • test_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the test split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • test_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the test split query, exclusive. Strings should be formatted in one of the following ormats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • storage_connector Optional[hsfs.StorageConnector]: Storage connector defining the sink location for the training dataset, defaults to None, and materializes training dataset on HopsFS.
  • location Optional[str]: Path to complement the sink storage connector with, e.g if the storage connector points to an S3 bucket, this path can be used to define a sub-directory inside the bucket to place the training dataset. Defaults to "", saving the training dataset at the root defined by the storage connector.
  • description Optional[str]: A string describing the contents of the training dataset to improve discoverability for Data Scientists, defaults to empty string "".
  • extra_filter Optional[Union[hsfs.constructor.filter.Filter, hsfs.constructor.filter.Logic]]: Additional filters to be attached to the training dataset. The filters will be also applied in get_batch_data.
  • data_format Optional[str]: The data format used to save the training dataset, defaults to "parquet"-format.
  • coalesce Optional[bool]: If true the training dataset data will be coalesced into a single partition before writing. The resulting training dataset will be a single file per split. Default False.
  • seed Optional[int]: Optionally, define a seed to create the random splits with, in order to guarantee reproducability, defaults to None.
  • statistics_config Optional[Union[hsfs.StatisticsConfig, bool, dict]]: A configuration object, or a dictionary with keys "enabled" to generally enable descriptive statistics computation for this feature group, "correlations" to turn on feature correlation computation and "histograms" to compute feature value frequencies. The values should be booleans indicating the setting. To fully turn off statistics computation pass statistics_config=False. Defaults to None and will compute only descriptive statistics.
  • write_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. When using the python engine, write_options can contain the following entries:
    • key spark and value an object of type hsfs.core.job_configuration.JobConfiguration to configure the Hopsworks Job used to compute the training dataset.
    • key wait_for_job and value True or False to configure whether or not to the save call should return only after the Hopsworks Job has finished. By default it waits. Defaults to {}.

Returns

(td_version, Job): Tuple of training dataset version and job. When using the python engine, it returns the Hopsworks Job that was launched to create the training dataset.

[source]

create_train_validation_test_split#

FeatureView.create_train_validation_test_split(
    validation_size=None,
    test_size=None,
    train_start="",
    train_end="",
    validation_start="",
    validation_end="",
    test_start="",
    test_end="",
    storage_connector=None,
    location="",
    description="",
    extra_filter=None,
    data_format="parquet",
    coalesce=False,
    seed=None,
    statistics_config=None,
    write_options={},
)

Create the metadata for a training dataset and save the corresponding training data into location. The training data is split into train, validation, and test set at random or according to time range. The training data can be retrieved by calling feature_view.get_train_validation_test_split.

Create random splits

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# create a train-validation-test split dataset
version, job = feature_view.create_train_validation_test_split(
    validation_size=0.3,
    test_size=0.2,
    description='Description of a dataset',
    data_format='csv'
)

Create time series splits by specifying date as string

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up dates
train_start = "2022-01-01 00:00:00"
train_end = "2022-06-01 23:59:59"
validation_start = "2022-06-02 00:00:00"
validation_end = "2022-07-01 23:59:59"
test_start = "2022-07-02 00:00:00"
test_end = "2022-08-01 23:59:59"

# create a train-validation-test split dataset
version, job = feature_view.create_train_validation_test_split(
    train_start=train_start,
    train_end=train_end,
    validation_start=validation_start,
    validation_end=validation_end,
    test_start=test_start,
    test_end=test_end,
    description='Description of a dataset',
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format='csv'
)

Create time series splits by specifying date as datetime object

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up dates
from datetime import datetime
date_format = "%Y-%m-%d %H:%M:%S"

train_start = datetime.strptime("2022-01-01 00:00:00", date_format)
train_end = datetime.strptime("2022-06-06 23:59:59", date_format)
validation_start = datetime.strptime("2022-06-02 00:00:00", date_format)
validation_end = datetime.strptime("2022-07-01 23:59:59", date_format)
test_start = datetime.strptime("2022-06-07 00:00:00", date_format)
test_end = datetime.strptime("2022-12-25 23:59:59", date_format)

# create a train-validation-test split dataset
version, job = feature_view.create_train_validation_test_split(
    train_start=train_start,
    train_end=train_end,
    validation_start=validation_start,
    validation_end=validation_end,
    test_start=test_start,
    test_end=test_end,
    description='Description of a dataset',
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format='csv'
)

Write training dataset to external storage

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get storage connector instance
external_storage_connector = fs.get_storage_connector("storage_connector_name")

# create a train-validation-test split dataset
version, job = feature_view.create_train_validation_test_split(
    train_start=...,
    train_end=...,
    validation_start=...,
    validation_end=...,
    test_start=...,
    test_end=...,
    description=...,
    storage_connector = external_storage_connector,
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format=...
)

Data Formats

The feature store currently supports the following data formats for training datasets:

  1. tfrecord
  2. csv
  3. tsv
  4. parquet
  5. avro
  6. orc

Currently not supported petastorm, hdf5 and npy file formats.

Arguments

  • validation_size Optional[float]: size of validation set.
  • test_size Optional[float]: size of test set.
  • train_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the train split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • train_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the train split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • validation_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the validation split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • validation_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the validation split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • test_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the test split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • test_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the test split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • storage_connector Optional[hsfs.StorageConnector]: Storage connector defining the sink location for the training dataset, defaults to None, and materializes training dataset on HopsFS.
  • location Optional[str]: Path to complement the sink storage connector with, e.g if the storage connector points to an S3 bucket, this path can be used to define a sub-directory inside the bucket to place the training dataset. Defaults to "", saving the training dataset at the root defined by the storage connector.
  • description Optional[str]: A string describing the contents of the training dataset to improve discoverability for Data Scientists, defaults to empty string "".
  • extra_filter Optional[Union[hsfs.constructor.filter.Filter, hsfs.constructor.filter.Logic]]: Additional filters to be attached to the training dataset. The filters will be also applied in get_batch_data.
  • data_format Optional[str]: The data format used to save the training dataset, defaults to "parquet"-format.
  • coalesce Optional[bool]: If true the training dataset data will be coalesced into a single partition before writing. The resulting training dataset will be a single file per split. Default False.
  • seed Optional[int]: Optionally, define a seed to create the random splits with, in order to guarantee reproducability, defaults to None.
  • statistics_config Optional[Union[hsfs.StatisticsConfig, bool, dict]]: A configuration object, or a dictionary with keys "enabled" to generally enable descriptive statistics computation for this feature group, "correlations" to turn on feature correlation computation and "histograms" to compute feature value frequencies. The values should be booleans indicating the setting. To fully turn off statistics computation pass statistics_config=False. Defaults to None and will compute only descriptive statistics.
  • write_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. When using the python engine, write_options can contain the following entries:
    • key spark and value an object of type hsfs.core.job_configuration.JobConfiguration to configure the Hopsworks Job used to compute the training dataset.
    • key wait_for_job and value True or False to configure whether or not to the save call should return only after the Hopsworks Job has finished. By default it waits. Defaults to {}.

Returns

(td_version, Job): Tuple of training dataset version and job. When using the python engine, it returns the Hopsworks Job that was launched to create the training dataset.

[source]

create_training_data#

FeatureView.create_training_data(
    start_time="",
    end_time="",
    storage_connector=None,
    location="",
    description="",
    extra_filter=None,
    data_format="parquet",
    coalesce=False,
    seed=None,
    statistics_config=None,
    write_options={},
)

Create the metadata for a training dataset and save the corresponding training data into location. The training data can be retrieved by calling feature_view.get_training_data.

Create training dataset

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# create a training dataset
version, job = feature_view.create_training_data(
    description='Description of a dataset',
    data_format='csv',
    # async creation in order not to wait till finish of the job
    write_options={"wait_for_job": False}
)

Create training data specifying date range with dates as strings

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up dates
start_time = "2022-01-01 00:00:00"
end_time = "2022-06-06 23:59:59"

# create a training dataset
version, job = feature_view.create_training_data(
    start_time=start_time,
    end_time=end_time,
    description='Description of a dataset',
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format='csv'
)

# When we want to read the training data, we need to supply the training data version returned by the create_training_data method:
X_train, X_test, y_train, y_test = feature_view.get_training_data(version)

Create training data specifying date range with dates as datetime objects

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up dates
from datetime import datetime
date_format = "%Y-%m-%d %H:%M:%S"

start_time = datetime.strptime("2022-01-01 00:00:00", date_format)
end_time = datetime.strptime("2022-06-06 23:59:59", date_format)

# create a training dataset
version, job = feature_view.create_training_data(
    start_time=start_time,
    end_time=end_time,
    description='Description of a dataset',
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format='csv'
)

Write training dataset to external storage

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get storage connector instance
external_storage_connector = fs.get_storage_connector("storage_connector_name")

# create a train-test split dataset
version, job = feature_view.create_training_data(
    start_time=...,
    end_time=...,
    storage_connector = external_storage_connector,
    description=...,
    # you can have different data formats such as csv, tsv, tfrecord, parquet and others
    data_format=...
)

Data Formats

The feature store currently supports the following data formats for training datasets:

  1. tfrecord
  2. csv
  3. tsv
  4. parquet
  5. avro
  6. orc

Currently not supported petastorm, hdf5 and npy file formats.

Arguments

  • start_time Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the training dataset query, inclusive. Optional. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • end_time Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the training dataset query, exclusive. Optional. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • storage_connector Optional[hsfs.StorageConnector]: Storage connector defining the sink location for the training dataset, defaults to None, and materializes training dataset on HopsFS.
  • location Optional[str]: Path to complement the sink storage connector with, e.g if the storage connector points to an S3 bucket, this path can be used to define a sub-directory inside the bucket to place the training dataset. Defaults to "", saving the training dataset at the root defined by the storage connector.
  • description Optional[str]: A string describing the contents of the training dataset to improve discoverability for Data Scientists, defaults to empty string "".
  • extra_filter Optional[Union[hsfs.constructor.filter.Filter, hsfs.constructor.filter.Logic]]: Additional filters to be attached to the training dataset. The filters will be also applied in get_batch_data.
  • data_format Optional[str]: The data format used to save the training dataset, defaults to "parquet"-format.
  • coalesce Optional[bool]: If true the training dataset data will be coalesced into a single partition before writing. The resulting training dataset will be a single file per split. Default False.
  • seed Optional[int]: Optionally, define a seed to create the random splits with, in order to guarantee reproducability, defaults to None.
  • statistics_config Optional[Union[hsfs.StatisticsConfig, bool, dict]]: A configuration object, or a dictionary with keys "enabled" to generally enable descriptive statistics computation for this feature group, "correlations" to turn on feature correlation computation and "histograms" to compute feature value frequencies. The values should be booleans indicating the setting. To fully turn off statistics computation pass statistics_config=False. Defaults to None and will compute only descriptive statistics.
  • write_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. When using the python engine, write_options can contain the following entries:
    • key spark and value an object of type hsfs.core.job_configuration.JobConfiguration to configure the Hopsworks Job used to compute the training dataset.
    • key wait_for_job and value True or False to configure whether or not to the save call should return only after the Hopsworks Job has finished. By default it waits. Defaults to {}.

Returns

(td_version, Job): Tuple of training dataset version and job. When using the python engine, it returns the Hopsworks Job that was launched to create the training dataset.

[source]

delete#

FeatureView.delete()

Delete current feature view, all associated metadata and training data.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# delete a feature view
feature_view.delete()

Potentially dangerous operation

This operation drops all metadata associated with this version of the feature view and related training dataset and materialized data in HopsFS.

Raises

hsfs.client.exceptions.RestAPIError.

[source]

delete_all_training_datasets#

FeatureView.delete_all_training_datasets()

Delete all training datasets. This will delete both metadata and training data.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# delete all training datasets
feature_view.delete_all_training_datasets()

Raises

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

[source]

delete_tag#

FeatureView.delete_tag(name)

Delete a tag attached to a feature view.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# delete a tag
feature_view.delete_tag('name_of_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]

delete_training_dataset#

FeatureView.delete_training_dataset(training_dataset_version)

Delete a training dataset. This will delete both metadata and training data.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# delete a training dataset
feature_view.delete_training_dataset(
    training_dataset_version=1
)

Arguments

  • training_dataset_version int: Version of the training dataset to be removed.

Raises

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

[source]

delete_training_dataset_tag#

FeatureView.delete_training_dataset_tag(training_dataset_version, name)

Delete a tag attached to a training dataset.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# delete training dataset tag
feature_view.delete_training_dataset_tag(
    training_dataset_version=1,
    name='name_of_dataset'
)

Arguments

  • training_dataset_version int: training dataset version
  • name str: Name of the tag to be removed.

Raises

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

[source]

from_response_json#

FeatureView.from_response_json(json_dict)

[source]

get_batch_data#

FeatureView.get_batch_data(start_time=None, end_time=None, read_options=None)

Get a batch of data from an event time interval from the offline feature store.

Batch data for the last 24 hours

    # get feature store instance
    fs = ...

    # get feature view instance
    feature_view = fs.get_feature_view(...)

    # set up dates
    import datetime
    start_date = (datetime.datetime.now() - datetime.timedelta(hours=24))
    end_date = (datetime.datetime.now())

    # get a batch of data
    df = feature_view.get_batch_data(
        start_time=start_date,
        end_time=end_date
    )

Arguments

  • start_time Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the batch query, inclusive. Optional. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • end_time Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the batch query, exclusive. Optional. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • read_options: User provided read options. Defaults to {}.

Returns

DataFrame: A dataframe

[source]

get_batch_query#

FeatureView.get_batch_query(start_time=None, end_time=None)

Get a query string of the batch query.

Batch query for the last 24 hours

    # get feature store instance
    fs = ...

    # get feature view instance
    feature_view = fs.get_feature_view(...)

    # set up dates
    import datetime
    start_date = (datetime.datetime.now() - datetime.timedelta(hours=24))
    end_date = (datetime.datetime.now())

    # get a query string of batch query
    query_str = feature_view.get_batch_query(
        start_time=start_date,
        end_time=end_date
    )
    # print query string
    print(query_str)

Arguments

  • start_time Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the batch query, inclusive. Optional. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • end_time Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the batch query, exclusive. Optional. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.

Returns

str: batch query

[source]

get_feature_vector#

FeatureView.get_feature_vector(entry, passed_features={}, external=None)

Returns assembled feature vector from online feature store. Call feature_view.init_serving before this method if the following configurations are needed. 1. The training dataset version of the transformation statistics 2. Additional configurations of online serving engine

Missing primary key entries

If the provided primary key entry can't be found in one or more of the feature groups used by this feature view the call to this method will raise an exception.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get a feature vector
feature_view.get_feature_vector(
    entry = {"pk1": 1, "pk2": 2}
)

Get feature vector with user-supplied features

# get feature store instance
fs = ...
# get feature view instance
feature_view = fs.get_feature_view(...)

# the application provides a feature value 'app_attr'
app_attr = ...

# get a feature vector
feature_view.get_feature_vector(
    entry = {"pk1": 1, "pk2": 2},
    passed_features = { "app_feature" : app_attr }
)

Arguments

  • entry Dict[str, Any]: dictionary of feature group primary key and values provided by serving application.
  • passed_features Optional[Dict[str, Any]]: dictionary of feature values provided by the application at runtime. They can replace features values fetched from the feature store as well as providing feature values which are not available in the feature store.
  • external Optional[bool]: boolean, optional. If set to True, the connection to the online feature store is established using the same host as for the host parameter in the hsfs.connection() method. If set to False, the online feature store storage connector is used which relies on the private IP. Defaults to True if connection to Hopsworks is established from external environment (e.g AWS Sagemaker or Google Colab), otherwise to False.

Returns

list List of feature values related to provided primary keys, ordered according to positions of this features in the feature view query.

Raises

Exception. When primary key entry cannot be found in one or more of the feature groups used by this feature view.

[source]

get_feature_vectors#

FeatureView.get_feature_vectors(entry, passed_features={}, external=None)

Returns assembled feature vectors in batches from online feature store. Call feature_view.init_serving before this method if the following configurations are needed. 1. The training dataset version of the transformation statistics 2. Additional configurations of online serving engine

Missing primary key entries

If any of the provided primary key elements in entry can't be found in any of the feature groups, no feature vector for that primary key value will be returned. If it can be found in at least one but not all feature groups used by this feature view the call to this method will raise an exception.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get assembled serving vectors
feature_view.get_feature_vectors(
    entry = [
        {"pk1": 1, "pk2": 2},
        {"pk1": 3, "pk2": 4},
        {"pk1": 5, "pk2": 6}
    ]
)

Arguments

  • entry List[Dict[str, Any]]: a list of dictionary of feature group primary key and values provided by serving application.
  • passed_features Optional[List[Dict[str, Any]]]: a list of dictionary of feature values provided by the application at runtime. They can replace features values fetched from the feature store as well as providing feature values which are not available in the feature store.
  • external Optional[bool]: boolean, optional. If set to True, the connection to the online feature store is established using the same host as for the host parameter in the hsfs.connection() method. If set to False, the online feature store storage connector is used which relies on the private IP. Defaults to True if connection to Hopsworks is established from external environment (e.g AWS Sagemaker or Google Colab), otherwise to False.

Returns

List[list] List of lists of feature values related to provided primary keys, ordered according to positions of this features in the feature view query.

Raises

Exception. When primary key entry cannot be found in one or more of the feature groups used by this feature view.

[source]

get_parent_feature_groups#

FeatureView.get_parent_feature_groups()

Get the parents of this feature view, 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.

[source]

get_tag#

FeatureView.get_tag(name)

Get the tags of a feature view.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get a tag of a feature view
name = feature_view.get_tag('tag_name')

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#

FeatureView.get_tags()

Returns all tags attached to a training dataset.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get tags
list_tags = feature_view.get_tags()

Returns

Dict[str, obj] of tags.

Raises

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

[source]

get_train_test_split#

FeatureView.get_train_test_split(training_dataset_version, read_options=None)

Get training data created by feature_view.create_train_test_split or feature_view.train_test_split.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get training data
X_train, X_test, y_train, y_test = feature_view.get_train_test_split(training_dataset_version=1)

Arguments

  • training_dataset_version: training dataset version
  • read_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. For python engine:
    • key "hive_config" to pass a dictionary of hive or tez configurations. For example: {"hive_config": {"hive.tez.cpu.vcores": 2, "tez.grouping.split-count": "3"}} Defaults to {}.

Returns

(X_train, X_test, y_train, y_test): Tuple of dataframe of features and labels

[source]

get_train_validation_test_split#

FeatureView.get_train_validation_test_split(training_dataset_version, read_options=None)

Get training data created by feature_view.create_train_validation_test_split or feature_view.train_validation_test_split.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get training data
X_train, X_val, X_test, y_train, y_val, y_test = feature_view.get_train_validation_test_splits(training_dataset_version=1)

Arguments

  • training_dataset_version: training dataset version
  • read_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. For python engine:
    • key "hive_config" to pass a dictionary of hive or tez configurations. For example: {"hive_config": {"hive.tez.cpu.vcores": 2, "tez.grouping.split-count": "3"}} Defaults to {}.

Returns

(X_train, X_val, X_test, y_train, y_val, y_test): Tuple of dataframe of features and labels

[source]

get_training_data#

FeatureView.get_training_data(training_dataset_version, read_options=None)

Get training data created by feature_view.create_training_data or feature_view.training_data.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get training data
features_df, labels_df = feature_view.get_training_data(training_dataset_version=1)

External Storage Support

Reading training data that was written to external storage using a Storage Connector other than S3 can currently not be read using HSFS APIs with Python as Engine, instead you will have to use the storage's native client.

Arguments

  • training_dataset_version: training dataset version
  • read_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. For python engine:
    • key "hive_config" to pass a dictionary of hive or tez configurations. For example: {"hive_config": {"hive.tez.cpu.vcores": 2, "tez.grouping.split-count": "3"}} Defaults to {}.

Returns

(X, y): Tuple of dataframe of features and labels

[source]

get_training_dataset_tag#

FeatureView.get_training_dataset_tag(training_dataset_version, name)

Get the tags of a training dataset.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get a training dataset tag
tag_str = feature_view.get_training_dataset_tag(
    training_dataset_version=1,
     name="tag_schema"
)

Arguments

  • training_dataset_version int: training dataset version
  • 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_training_dataset_tags#

FeatureView.get_training_dataset_tags(training_dataset_version)

Returns all tags attached to a training dataset.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get a training dataset tags
list_tags = feature_view.get_training_dataset_tags(
    training_dataset_version=1
)

Returns

Dict[str, obj] of tags.

Raises

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

[source]

init_batch_scoring#

FeatureView.init_batch_scoring(training_dataset_version=None)

Initialise feature view to retrieve feature vector from offline feature store.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# initialise feature view to retrieve feature vector from offline feature store
feature_view.init_batch_scoring(training_dataset_version=1)

# get batch data
batch_data = feature_view.get_batch_data(...)

Arguments

  • training_dataset_version Optional[int]: int, optional. Default to be None. Transformation statistics are fetched from training dataset and applied to the feature vector.

[source]

init_serving#

FeatureView.init_serving(training_dataset_version=None, external=None, options=None)

Initialise feature view to retrieve feature vector from online feature store.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# initialise feature view to retrieve a feature vector
feature_view.init_serving(training_dataset_version=1)

Arguments

  • training_dataset_version Optional[int]: int, optional. Default to be 1. Transformation statistics are fetched from training dataset and applied to the feature vector.
  • external Optional[bool]: boolean, optional. If set to True, the connection to the online feature store is established using the same host as for the host parameter in the hsfs.connection() method. If set to False, the online feature store storage connector is used which relies on the private IP. Defaults to True if connection to Hopsworks is established from external environment (e.g AWS Sagemaker or Google Colab), otherwise to False.
  • options Optional[dict]: Additional options as key/value pairs for configuring online serving engine.
    • key: kwargs of SqlAlchemy engine creation (See: https://docs.sqlalchemy.org/en/20/core/engines.html#sqlalchemy.create_engine). For example: {"pool_size": 10}

[source]

json#

FeatureView.json()

[source]

purge_all_training_data#

FeatureView.purge_all_training_data()

Delete all training datasets (data only).

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# purge all training data
feature_view.purge_all_training_data()

Raises

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

[source]

purge_training_data#

FeatureView.purge_training_data(training_dataset_version)

Delete a training dataset (data only).

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# purge training data
feature_view.purge_training_data(training_dataset_version=1)

Arguments

  • training_dataset_version int: Version of the training dataset to be removed.

Raises

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

[source]

recreate_training_dataset#

FeatureView.recreate_training_dataset(training_dataset_version, write_options=None)

Recreate a training dataset.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# recreate a training dataset that has been deleted
feature_view.recreate_training_dataset(training_dataset_version=1)

Info

If a materialised training data has deleted. Use recreate_training_dataset() to recreate the training data.

Arguments

  • training_dataset_version int: training dataset version
  • read_options: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. When using the python engine, write_options can contain the following entries:
    • key spark and value an object of type hsfs.core.job_configuration.JobConfiguration to configure the Hopsworks Job used to compute the training dataset.
    • key wait_for_job and value True or False to configure whether or not to the save call should return only after the Hopsworks Job has finished. By default it waits. Defaults to {}.

Returns

Job: When using the python engine, it returns the Hopsworks Job that was launched to create the training dataset.

[source]

to_dict#

FeatureView.to_dict()

[source]

train_test_split#

FeatureView.train_test_split(
    test_size=None,
    train_start="",
    train_end="",
    test_start="",
    test_end="",
    description="",
    extra_filter=None,
    statistics_config=None,
    read_options=None,
)

Create the metadata for a training dataset and get the corresponding training data from the offline feature store. This returns the training data in memory and does not materialise data in storage. The training data is split into train and test set at random or according to time ranges. The training data can be recreated by calling feature_view.get_train_test_split with the metadata created.

Create random train/test splits

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get training data
X_train, X_test, y_train, y_test = feature_view.train_test_split(
    test_size=0.2
)

Create time-series train/test splits

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up dates
train_start = "2022-05-01 00:00:00"
train_end = "2022-06-04 23:59:59"
test_start = "2022-07-01 00:00:00"
test_end= "2022-08-04 23:59:59"
# you can also pass dates as datetime objects

# get training data
X_train, X_test, y_train, y_test = feature_view.train_test_split(
    train_start=train_start,
    train_end=train_end,
    test_start=test_start,
    test_end=test_end,
    description='Description of a dataset'
)

Arguments

  • test_size Optional[float]: size of test set. Should be between 0 and 1.
  • train_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the train split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f.
  • train_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the train split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • test_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the test split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • test_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the test split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • description Optional[str]: A string describing the contents of the training dataset to improve discoverability for Data Scientists, defaults to empty string "".
  • extra_filter Optional[Union[hsfs.constructor.filter.Filter, hsfs.constructor.filter.Logic]]: Additional filters to be attached to the training dataset. The filters will be also applied in get_batch_data.
  • statistics_config Optional[Union[hsfs.StatisticsConfig, bool, dict]]: A configuration object, or a dictionary with keys "enabled" to generally enable descriptive statistics computation for this feature group, "correlations" to turn on feature correlation computation and "histograms" to compute feature value frequencies. The values should be booleans indicating the setting. To fully turn off statistics computation pass statistics_config=False. Defaults to None and will compute only descriptive statistics.
  • read_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. When using the python engine, write_options can contain the following entries:
    • key "hive_config" to pass a dictionary of hive or tez configurations. For example: {"hive_config": {"hive.tez.cpu.vcores": 2, "tez.grouping.split-count": "3"}}
    • key spark and value an object of type hsfs.core.job_configuration.JobConfiguration to configure the Hopsworks Job used to compute the training dataset. Defaults to {}.

Returns

(X_train, X_test, y_train, y_test): Tuple of dataframe of features and labels

[source]

train_validation_test_split#

FeatureView.train_validation_test_split(
    validation_size=None,
    test_size=None,
    train_start="",
    train_end="",
    validation_start="",
    validation_end="",
    test_start="",
    test_end="",
    description="",
    extra_filter=None,
    statistics_config=None,
    read_options=None,
)

Create the metadata for a training dataset and get the corresponding training data from the offline feature store. This returns the training data in memory and does not materialise data in storage. The training data is split into train, validation, and test set at random or according to time ranges. The training data can be recreated by calling feature_view.get_train_validation_test_split with the metadata created.

Example

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get training data
X_train, X_val, X_test, y_train, y_val, y_test = feature_view.train_validation_test_split(
    validation_size=0.3,
    test_size=0.2
)

Time Series split

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up dates
start_time_train = '2017-01-01 00:00:01'
end_time_train = '2018-02-01 23:59:59'

start_time_val = '2018-02-02 23:59:59'
end_time_val = '2019-02-01 23:59:59'

start_time_test = '2019-02-02 23:59:59'
end_time_test = '2020-02-01 23:59:59'
# you can also pass dates as datetime objects

# get training data
X_train, X_val, X_test, y_train, y_val, y_test = feature_view.train_validation_test_split(
    train_start=start_time_train,
    train_end=end_time_train,
    validation_start=start_time_val,
    validation_end=end_time_val,
    test_start=start_time_test,
    test_end=end_time_test
)

Arguments

  • validation_size Optional[float]: size of validation set. Should be between 0 and 1.
  • test_size Optional[float]: size of test set. Should be between 0 and 1.
  • train_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the train split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • train_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the train split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • validation_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the validation split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • validation_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the validation split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • test_start Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the test split query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • test_end Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the test split query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • description Optional[str]: A string describing the contents of the training dataset to improve discoverability for Data Scientists, defaults to empty string "".
  • extra_filter Optional[Union[hsfs.constructor.filter.Filter, hsfs.constructor.filter.Logic]]: Additional filters to be attached to the training dataset. The filters will be also applied in get_batch_data.
  • statistics_config Optional[Union[hsfs.StatisticsConfig, bool, dict]]: A configuration object, or a dictionary with keys "enabled" to generally enable descriptive statistics computation for this feature group, "correlations" to turn on feature correlation computation and "histograms" to compute feature value frequencies. The values should be booleans indicating the setting. To fully turn off statistics computation pass statistics_config=False. Defaults to None and will compute only descriptive statistics.
  • read_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. When using the python engine, write_options can contain the following entries:
    • key "hive_config" to pass a dictionary of hive or tez configurations. For example: {"hive_config": {"hive.tez.cpu.vcores": 2, "tez.grouping.split-count": "3"}}
    • key spark and value an object of type hsfs.core.job_configuration.JobConfiguration to configure the Hopsworks Job used to compute the training dataset. Defaults to {}.

Returns

(X_train, X_val, X_test, y_train, y_val, y_test): Tuple of dataframe of features and labels

[source]

training_data#

FeatureView.training_data(
    start_time=None,
    end_time=None,
    description="",
    extra_filter=None,
    statistics_config=None,
    read_options=None,
)

Create the metadata for a training dataset and get the corresponding training data from the offline feature store. This returns the training data in memory and does not materialise data in storage. The training data can be recreated by calling feature_view.get_training_data with the metadata created.

Create random splits

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# get training data
features_df, labels_df  = feature_view.training_data(
    description='Descriprion of a dataset',
)

Create time-series based splits

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

# set up a date
start_time = "2022-05-01 00:00:00"
end_time = "2022-06-04 23:59:59"
# you can also pass dates as datetime objects

# get training data
features_df, labels_df = feature_view.training_data(
    start_time=start_time,
    end_time=end_time,
    description='Description of a dataset'
)

Arguments

  • start_time Optional[Union[str, int, datetime.datetime, datetime.date]]: Start event time for the training dataset query, inclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • end_time Optional[Union[str, int, datetime.datetime, datetime.date]]: End event time for the training dataset query, exclusive. Strings should be formatted in one of the following formats %Y-%m-%d, %Y-%m-%d %H, %Y-%m-%d %H:%M, %Y-%m-%d %H:%M:%S, or %Y-%m-%d %H:%M:%S.%f. Int, i.e Unix Epoch should be in seconds.
  • description Optional[str]: A string describing the contents of the training dataset to improve discoverability for Data Scientists, defaults to empty string "".
  • extra_filter Optional[Union[hsfs.constructor.filter.Filter, hsfs.constructor.filter.Logic]]: Additional filters to be attached to the training dataset. The filters will be also applied in get_batch_data.
  • statistics_config Optional[Union[hsfs.StatisticsConfig, bool, dict]]: A configuration object, or a dictionary with keys "enabled" to generally enable descriptive statistics computation for this feature group, "correlations" to turn on feature correlation computation and "histograms" to compute feature value frequencies. The values should be booleans indicating the setting. To fully turn off statistics computation pass statistics_config=False. Defaults to None and will compute only descriptive statistics.
  • read_options Optional[Dict[Any, Any]]: Additional options as key/value pairs to pass to the execution engine. For spark engine: Dictionary of read options for Spark. When using the python engine, write_options can contain the following entries:
    • key "hive_config" to pass a dictionary of hive or tez configurations. For example: {"hive_config": {"hive.tez.cpu.vcores": 2, "tez.grouping.split-count": "3"}}
    • key spark and value an object of type hsfs.core.job_configuration.JobConfiguration to configure the Hopsworks Job used to compute the training dataset. Defaults to {}.

Returns

(X, y): Tuple of dataframe of features and labels. If there are no labels, y returns None.

[source]

update#

FeatureView.update()

Update the description of the feature view.

Update the feature view with a new description.

# get feature store instance
fs = ...

# get feature view instance
feature_view = fs.get_feature_view(...)

feature_view.description = "new description"
feature_view.update()

# Description is updated in the metadata. Below should return "new description".
fs.get_feature_view("feature_view_name", 1).description

Returns

FeatureView Updated feature view.

Raises

hsfs.client.exceptions.RestAPIError.

[source]

update_from_response_json#

FeatureView.update_from_response_json(json_dict)