Entities

Entities play vital roles in:

1. Defining and Storing Features and Labels: The featureform library provides an entity decorator that can be applied to classes. This decorator allows you to associate features and labels with the entity using Featureform data sets that you’ve defined earlier. A feature typically comprises at least two columns: an entity column (similar to an index or primary key) and a value column. Optionally, it can include a third column for timestamps, which becomes essential for constructing point-in-time correct training sets.

   For example, consider a “User” entity with an “age” feature and a “credit score” label associated with it:

   import featureform as ff
   
   @ff.entity
   class User:
     age = ff.Feature(dataset[["user", "age"]], variant="simple", type=ff.Int, inference_store=redis)
     credit_score = ff.Label(dataset[["user", "credit_score"]], variant="simple", type=ff.Int)
   

2. Building Training Sets by Joining Features and Labels by Their Entity: To create a training set, you need to associate a label with a set of features based on their entity values. If both the label and features have timestamps, the feature value chosen is the one with a matching entity and a timestamp that is the closest (but earlier) to the timestamp of the label.

   In the case of our “User” entity example, we can create a training set like this:

   ff.register_training_set(
     "credit_score_training_set", "v1",
     label=("credit_score", "simple"),
     features=[("age", "simple")],
   )
   

   Alternatively, if the “User” class is in scope, you can use a convenient syntax to avoid explicitly specifying name/variant tuples:

   ff.register_training_set(
     "credit_score_training_set", "v1",
     label=User.credit_score,
     features=User.age,
   )
   

3. Serving Features at Inference Time by Entity: The inference store retains the most recent value of each feature, indexed by its entity column. To retrieve features, you can utilize the features method of a Featureform Client object.

   f = client.features([("age", "simple")], entities={"user": "10323232"})
   

Entities provide a structured and organized approach to managing features and labels in Featureform, facilitating the creation of training sets and the retrieval of features during inference.