ItamiForge
Projects

YAAML

Lightweight, production-ready AutoML built entirely on scikit-learn with Python 3.12+ modernization

YAAML (Yet Another AutoML) is an old attempt at creating a fully-automated machine learning package. It is built entirely on scikit-learn without complex dependencies, focusing on modernization and zero-bloat.

Overview

Designed to sit directly on top of sklearn, YAAML aims to streamline the modeling workflow through an intelligent YAAMLAutoML wrapper pipeline.

Key Capabilities

  • Dependency-Free Backbone: No massive deep learning libraries required—runs natively with scikit-learn.
  • Smart Data Preprocessing: Handles missing data imputation, target encoding, and multi-type dataframes.
  • Multi-Task Support: Automatically detects and adapts execution for classification vs regression.
  • Modern Python Validation: Fully typed with Python 3.12+ features, using union types and the walrus operator.

Architecture

The system treats feature engineering and model selection as a unified Search Space, utilizing hyperopt or randomized search internally to discover the best sequence of transformations and estimator configurations.

This provides a true end-to-end framework where categorical casting, target encoding, data scaling, and model hyperparameters are optimized holistically.

Source

Available on GitHub and built to handle complex, messy real-world datasets in a transparent sklearn pipeline object.