Skip to main content

Retrieving the Training data from Feature Store.

We can use the Offline Store to create a training dataset & train the models with it.

Let's create a Training Dataset.​

  • To keep the consistency between Training & Testing Datasets, we will use an entity dataframe which consists of id & event timestamp column.
  • By using this entity dataframe, feature store will find the historical records by doing point-in-time joins.
# Let's load the Entity Dataframe.

entity_df = pd.read_csv("entity_df.csv")

Note: Make sure that this dataset's data types are also accurate.

Retrieving Historical Features.​

# Now let's use this entity df to create a training dataset with the historical features.

train_data = fs.get_historical_features(
    entity_df = entity_df, # Your Entity Dataframe
    feature_view = ["default_loan_feature_view"], # The name that you gave to create your feature view.
    features = cols, # The features that we want to retrieve from offline store. 
).to_df() # We are directly converting it to a pandas Dataframe.

Your training dataset is ready. You can use the same to train your models.