mlforge-sdk 0.3.0

ML Platform for your local machine using cheap cloud services for scalable resources.

mlforge

A simple feature store SDK for machine learning workflows.

Installation

pip install mlforge-sdk

Or with uv:

uv add mlforge-sdk

Quick Start

Define features using the @feature decorator:

import polars as pl
from mlforge import feature, entity_key

# Create reusable entity key transforms
with_user_id = entity_key("first", "last", "dob", alias="user_id")

@feature(
    keys=["user_id"],
    source="data/transactions.parquet",
    description="Total spend by user"
)
def user_total_spend(df: pl.DataFrame) -> pl.DataFrame:
    return (
        df.pipe(with_user_id)
        .group_by("user_id")
        .agg(pl.col("amt").sum().alias("total_spend"))
    )

Register features and materialize them:

from mlforge import Definitions, LocalStore
import my_features

defs = Definitions(
    name="my-project",
    features=[my_features],
    offline_store=LocalStore("./feature_store")
)

# Materialize features to storage
defs.materialize()

Retrieve features for training:

from mlforge import get_training_data

training_df = get_training_data(
    features=["user_total_spend"],
    entity_df=transactions,
    entities=[with_user_id],
    timestamp="trans_date_trans_time"  # Point-in-time correct joins
)

Features

• Simple API: Define features with a @feature decorator
• Entity Keys: Generate surrogate keys from natural keys using entity_key()
• Local Storage: Persist features to Parquet with LocalStore
• Point-in-Time Joins: Retrieve training data with temporal correctness using get_training_data()
• CLI: Build and list features from the command line

CLI Usage

Build features:

mlforge build definitions.py

Build specific features:

mlforge build definitions.py --features user_total_spend,merchant_total_spend

List registered features:

mlforge list definitions.py

Documentation

Full documentation is available at https://chonalchendo.github.io/mlforge

Contributing

Contributions are welcome! Please see the repository for development setup and guidelines.

License

MIT License - see LICENSE for details.