Array Features

Array Features

Tecton now natively supports Array type features with Float32, Float64, Int64, and String type elements.

This allows Tecton users to compactly store and serve features like dense embeddings or lists of categorical values.

Here’s an example On-Demand Feature View Overview that uses array features to compute the cosine similarity between a precomputed user embedding and a query embedding:

from tecton import RequestDataSource, Input, on_demand_feature_view
from pyspark.sql.types import StructType, StructField, FloatType, ArrayType, DoubleType
from ads.features.feature_tables.user_embeddings import user_embeddings
import pandas


request_schema = StructType()
request_schema.add(StructField('query_embedding', ArrayType(FloatType())))
request = RequestDataSource(request_schema=request_schema)

output_schema = StructType()
output_schema.add(StructField('cosine_similarity', DoubleType()))


@on_demand_feature_view(
    inputs={
        'request': Input(request),
        'user_embedding': Input(user_embeddings)
    },
    mode='pandas',
    output_schema=output_schema,
    family='fraud',
    owner='jake@tecton.ai',
    tags={'release': 'production'},
    description="Computes the cosine similarity between a query embedding and a precomputed user embedding."
)
def user_query_embedding_similarity(request: pandas.DataFrame, user_embedding: pandas.DataFrame):
    import pandas as pd
    import numpy as np
    from numpy.linalg import norm

    @np.vectorize
    def cosine_similarity(a: np.ndarray, b: np.ndarray):
        # Handle the case where there is no precomputed user embedding.
        if a is None or b is None:
            return -1.0

        return np.dot(a, b)/(norm(a)*norm(b))

    df = pd.DataFrame()
    df["cosine_similarity"] = (
        cosine_similarity(user_embedding["user_embedding"], request["query_embedding"]).astype('float64')
    )

    return df

View this code on Github.