This repository contains a Python implementation of M-learning with shrinkage toward equal weights (STEW) regularization applied to Tetris, as used in the article:
Lichtenberg, J. M. & Şimşek, Ö. (2019). Regularization in directable environments with application to Tetris. Proceedings of the 36th International Conference on Machine Learning, in PMLR 97:3953-3962
Further implementation details and pseudo-code of M-learning are available in the Supplementary Material.
Install required Python packages via
pip install -r requirements.txt
The following command runs M-learning with STEW for seven iterations, evaluating the algorithm after iterations 1, 3, and 7.
python run_stew_test.py
Other regularization terms can be tested by setting the regularization parameter to "ridge", "nonnegative", "ols" (= no regularization), or "ew" (equal weights).