kamil271e/ml-pipe-tab

ML-workflow skeleton: prediction for tabular data

ml-pipe-tab

The idea of this project is to provide a robust and comprehensive skeleton for typical machine learning prediction tasks regarding tabular data. The pipeline will leverage mainly sklearn and include typical steps such as data wrangling, feature engineering, feature selection, model selection, training, and fine-tuning. It should serve as a set of util tools for kaggle competitions.

Sources

• Feature engineering
• Imbalanced classification

ml-pipe-tab's Issues

Use Optuna for HPO

Implement a more refined version of fine-tuning by leveraging Bayesian optimization algorithms. Include functionality for both classification and regression problems. Allow specification of time limit for this process.

Heterogenous ensembles fix

Change example demo code (stack regressors and classifiers) to use more diverse components that could potentially work as a "mixture of experts".