What it does Xcessiv is a web-based Python tool for building stacked ensembles — feeding predictions from multiple base learners into another model — plus hyperparameter search and model comparison. It wraps the scikit-learn API, tracks hundreds of model-hyperparameter combinations, and exports the final stack as a standalone Python file.

The interesting bit The project explicitly targets a pain point the README calls “impractical to apply outside of Kaggle competitions”: proper stacked generalization requires careful cross-validated meta-feature generation, and Xcessiv automates that plumbing while still letting you define data sources, metrics, and learners in raw Python code.

Key highlights

• Any scikit-learn-compatible model works as a base learner
• Task-queue architecture for parallel hyperparameter search across multiple cores
• Bayesian optimization for automated hyperparameter tuning
• Direct TPOT integration for automated pipeline construction
• Greedy forward model selection for automated ensemble construction
• Supports both cross-validated stacked generalization and holdout-set blending

Caveats

• The project is explicitly alpha and unstable; future versions may break backwards compatibility with existing project files
• Development appears early-stage; the README asks for community guidance on direction

Verdict Worth a look if you routinely stack models or want a web UI to manage large hyperparameter search grids. Skip it if you need production stability today or your ensembles are simple enough to hand-roll in a notebook.