rodrigo-arenas/Sklearn-genetic-opt — reverse-engineered prompt

Build me a Python package for scikit learn that gives me a smarter alternative to GridSearchCV and RandomizedSearchCV by using genetic algorithms to tune model settings. I want it to feel like a drop in tool, so the main search class should work with the usual fit, predict, and best params style API, and it should support cross validation, mixed parameter types like integer, continuous, and categorical values, and parallel runs.

I also want a feature selection tool that can search for the best subset of features with cross validation, so I can use it together with hyperparameter tuning in one reproducible workflow. Please include practical stopping controls like stopping after no improvement, a small improvement threshold, or a time limit. Save optimization history so I can inspect progress and generate simple plots of fitness and convergence. If possible, support warm starts from known good configs and optional experiment tracking with MLflow.

Please make it usable as a real open source library with docs, examples, and tests. Look up current scikit learn docs if you need to.