- Type
- Feature Store
On this page
Enhance your machine learning workflows by integrating Feast, a powerful feature store, with ZenML. This integration enables efficient management and serving of features for model training and inference, streamlining the ML pipeline process.
Features with ZenML
- Seamless integration of Feast feature stores into ZenML pipelines
- Access to historical feature data for model training
- Simplified feature retrieval and management within ML workflows
- Improved collaboration and reproducibility across teams
Main Features
- Unified serving of features to models for training and inference
- Scalable offline store for batch scoring and model training
- Integration with various data sources and platforms
- Centralized feature registry and versioning
How to use ZenML with Feast
# First register the feature store and a stack
# zenml feature-store register feast_store --flavor=feast --feast_repo="<PATH/TO/FEAST/REPO>"
# zenml stack register ... -f feast_store
from zenml import step, pipeline
from zenml.client import Client
@step
def get_historical_features(
entity_dict, features, full_feature_names=False
):
feature_store = Client().active_stack.feature_store
entity_df = pd.DataFrame.from_dict(entity_dict)
return feature_store.get_historical_features(
entity_df=entity_df,
features=features,
full_feature_names=full_feature_names,
)
@pipeline
def my_pipeline():
my_features = get_historical_features(
entity_dict={"driver_id": [1001, 1002, 1003]},
features=["driver_hourly_stats:conv_rate", "driver_hourly_stats:acc_rate"]
)
# use features in downstream steps
# also use the CLI for Feast metadata etc
# zenml feature-store feature get-entities
# zenml feature-store feature get-data-sources
# ...
