Serving

Features

Returns the feature values for the specified entities.

Examples:

    client = ff.Client()
    fpf = client.features([("avg_transactions", "quickstart")], {"user": "C1410926"})
    # Run features through model

Parameters:

Name	Type	Description	Default
features	(list[str, str], list[str])	List of Name Variant Tuples	required
entities	dict	Dictionary of entity name/value pairs	required

Returns:

Name	Type	Description
features	Array	An Numpy array of feature values in the order given by the inputs

Training Sets

Return an iterator that iterates through the specified training set.

Examples:

    client = ff.Client()
    dataset = client.training_set("fraud_training", "quickstart")
    training_dataset = dataset.repeat(10).shuffle(1000).batch(8)
    for feature_batch in training_dataset:
        # Train model

Parameters:

Name	Type	Description	Default
name	str	Name of training set to be retrieved	required
variant	str	Variant of training set to be retrieved	''

Returns:

Name	Type	Description
training_set	Dataset	A training set iterator

Train/Test Split (Beta)

(This functionality is currently only available for Clickhouse).

Splits an existing training set into training and testing iterators. The split is processed on the underlying provider and calculated at serving time.

Examples:

import featureform as ff
client = ff.Client()
train, test = client
    .training_set("fraud_training", "v1")
    .train_test_split(
        test_size=0.7,
        train_size=0.3,
        shuffle=True,
        random_state=None,
        batch_size=5
    )

for features, label in train:
    print(features)
    print(label)
    clf.partial_fit(features, label)

for features, label in test:
    print(features)
    print(label)
    clf.score(features, label)


# TRAIN OUTPUT
# np.array([
#   [1, 1, 3],
#   [5, 1, 2],
#   [7, 6, 5],
#   [8, 3, 3],
#   [5, 2, 2],
# ])
# np.array([2, 4, 2, 3, 4])
# np.array([
#   [3, 1, 2],
#   [5, 4, 5],
# ])
# np.array([6, 7])

# TEST OUTPUT
# np.array([
#   [5, 1, 3],
#   [4, 3, 1],
#   [6, 6, 7],
# ])
# np.array([4, 6, 7])

Parameters:

Name	Type	Description	Default
test_size	float	The ratio of test set size to train set size. Must be a value between 0 and 1. If excluded it will be the complement to the train_size. One of test_size or train_size must be specified.	0
train_size	float	The ratio of train set size to train set size. Must be a value between 0 and 1. If excluded it will be the complement to the test_size. One of test_size or train_size must be specified.	0
shuffle	bool	Whether to shuffle the dataset before splitting.	True
random_state	Optional[int]	A random state to shuffle the dataset. If None, the dataset will be shuffled randomly on every call. If >0, the value will be used a seed to create random shuffle that can be repeated if subsequent calls use the same seed.	None
batch_size	int	The size of the batch to return from the iterator. Must be greater than 0.	1

Returns:

Name	Type	Description
train	Iterator	An iterator for training values.
test	Iterator	An iterator for testing values.

Sources

Return a dataframe from a registered source or transformation

Example:

definitions.py

transactions_df = client.dataframe("transactions", "quickstart")

avg_user_transaction_df = transactions_df.groupby("CustomerID")["TransactionAmount"].mean()

Parameters:

Name	Type	Description	Default
source	Union[SourceRegistrar, SubscriptableTransformation, str]	The source or transformation to compute the dataframe from	required
variant	str	The source variant; can't be None if source is a string	None
limit	int	The maximum number of records to return; defaults to NO_RECORD_LIMIT	NO_RECORD_LIMIT
asynchronous	bool	Flag to determine whether the client should wait for resources to be in either a READY or FAILED state before returning. Defaults to False to ensure that newly registered resources are in a READY state prior to serving them as dataframes.	False

Returns:

Name	Type	Description
df	DataFrame	The dataframe computed from the source or transformation

Nearest Neighbors

Query the K nearest neighbors of a provider vector in the index of a registered feature variant

Example:

definitions.py

# Get the 5 nearest neighbors of the vector [0.1, 0.2, 0.3] in the index of the feature "my_feature" with variant "my_variant"
nearest_neighbors = client.nearest("my_feature", "my_variant", [0.1, 0.2, 0.3], 5)
print(nearest_neighbors) # prints a list of entities (e.g. ["entity1", "entity2", "entity3", "entity4", "entity5"])

Parameters:

Name	Type	Description	Default
feature	Union[FeatureColumnResource, tuple(str, str)]	Feature object or tuple of Feature name and variant	required
vector	List[float]	Query vector	required
k	int	Number of nearest neighbors to return	required

Resource Location

Returns the location of a registered resource. For SQL resources, it will return the table name and for file resources, it will return the file path.

Example:

definitions.py

transaction_location = client.location("transactions", "quickstart", ff.SOURCE)

Parameters:

Name	Type	Description	Default
source	Union[SourceRegistrar, SubscriptableTransformation, str]	The source or transformation to compute the dataframe from	required
variant	str	The source variant; can't be None if source is a string	None
resource_type	ResourceType	The type of resource; can be one of ff.SOURCE, ff.FEATURE, ff.LABEL, or ff.TRAINING_SET	None