Ondemand Feature

An Ondemand Feature is a function that can be stored and called at serving time to run arbitrary code.

Defining An Ondemand Feature

Ondemand features have three arguments that can be used to pass in parameters, entities, or a serving client to call other serving functions within itself.

Arguments:

• client: The same client that is used to retrieve the ondemand feature with client.features().
• params: A list of parameters that are passed in at serving time. This can be any datatype or structure.
• entities: A list of entities that are passed in at serving time. This is optional.

On Demand Feature decorator.

Examples

import featureform as ff

@ff.ondemand_feature()
def avg_user_transactions(client, params, entities):
    pass

Parameters:

Name	Type	Description	Default
variant	str	Name of variant	''
name	str	Name of source	''
owner	Union[str, UserRegistrar]	Owner	''
description	str	Description of on demand feature	''
tags	List[str]	Optional grouping mechanism for resources	[]
properties	dict	Optional grouping mechanism for resources	{}

Returns:

Name	Type	Description
decorator	OnDemandFeature	decorator

Examples

Simple Aggregations

Ondemand features can be used to run simple aggregations on values that are passed in as params. In this example, we will simply be adding two numbers together.

@ff.ondemand_feature(variant="quickstart")
def avg_user_transactions(client, params, entities):
    return params[0] + params[1]


features = client.features([("avg_user_transactions", "quickstart")], params=[1, 2])
print(features)
# [3]

Aggregating Features

Suppose we have a feature that is the sum of two other features. We can use an ondemand feature to aggregate these internally and return the result.

@ff.ondemand_feature(variant="quickstart")
def lot_area(client, params, entities):
    """Returns the area of a property in square feet."""
    dim = client.features([("lot_width", "rev_1"), ("lot_length", "rev_1")], entities)
    return dim[0] * dim[1]

# Returns the area of a property of property 1432.
features = client.features([("lot_area", "sqft")], {"property_id": 1432})
print(features)
# [6098]

Nested Ondemand Features

We can call Ondemand Features from within other Ondemand Features. Using the previous example, we can create an Ondemand feature that converts the area from square feet to square meters.

@ff.ondemand_feature(variant="quickstart")
def lot_area(client, params, entities):
    area = client.features([("lot_area", "sqft")], entities)
    return area[0]/10.764

# Returns the area of a property of property 1432 in square meters.
features = client.features([("lot_area", "m2")], {"property_id": 1432})
print(features)
# [566.5227]