from feast import FeatureStore

store = FeatureStore(".")

# Get historical features as Ray Dataset
ds = store.get_historical_features(
    entity_df=entity_df,
    features=["embeddings:embedding"],
).to_ray_dataset()  # ← Proposed new method

# Now use Ray Data API directly
ds = ds.map_batches(custom_processing)
ds = ds.repartition(100)
predictions = ds.map_batches(ModelInference, num_gpus=1)

class FeatureStore:
    
    def to_ray_dataset(
        self,
        source: Optional[DataSource] = None,
        feature_view: Optional[FeatureView] = None,
        start_date: Optional[datetime] = None,
        end_date: Optional[datetime] = None,
    ) -> ray.data.Dataset:
        """Convert Feast data source or feature view to Ray Dataset."""
        ...
    
class RetrievalJob:
    
    def to_ray_dataset(self) -> ray.data.Dataset:
        """Convert retrieval result to Ray Dataset instead of Pandas."""
        ...