erp12 / lexicase-random-forests

Lexicase Selection and Random Forests

How to run experiments.

1. Create and activate a virtual environment for python 3.
2. Install requirements.
3. Run the main LexicaseRF.py script

python -m venv venv/
source venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt
python lexicaseRF.py [method] [task]

Methods

Currently there is only one supported method to experiment with: filter.

These experiments are for using lexicsae selections to pick the trees in a random forest.

Tasks

The supported tasks are classification and regression.

Caching training data and results

Training data and results will be stored in the data/ folder for faster reading during subsequent runs.

A results.csv file will be generated in the data/ folder which can be used to analyze results without rerunning experiments. To use the results.csv from previous experiements, add the --skip-train flag when running the script.

python lexicaseRF.py --skip-train

ToDo:

• Prevent lexicase from selecting the same tree twice?
• Grid search n_estimators, depth, and overpopulate_multiplier.
• Other related experiments mentioned on the push-language discourse