Evolution Strategies and Decision Transformers

In this repository, there is a codebase (codebase directory) and data from experiments (data directory) for paper Utilizing Evolution Strategies to Train Transformers in Reinforcement Learning by Matyáš Lorenc.

More information about the algorithms, the architecture, the experiments performed, and even the motivation behind this work can be found in the text of the article.

Directory codebase

Here, we can find our implementation of the OpenAI-ES and scripts carrying out the training (train_*.py) of a given model (feed-forward or decision transformer), the scripts enabling us to simulate the trained agents (play_*.py), and a script to plot the resulting logs (plot_experiment.py).

Directory data

Here in this directory, we can find results of the experiments. This means trained models, log files and plots.