File Description

• rl_hw3_demo.py
  This file contains the script for the expert algorithm used to collect training data, generating datasets based on expert strategies.

• dataset_episode_*.npz
  These files are the collected training datasets, with each file corresponding to a dataset of different size or quality. The specific reward results are shown in the figure below:
  

• environment.yml
  Please use the following recommended environment setup.

  Behavior Cloning and CQL Algorithm Implementation

This repository contains the implementation of Behavior Cloning (BC) and Conservative Q-Learning (CQL) algorithms for training reinforcement learning models using a dataset of state-action pairs. It provides an environment for experimenting with these algorithms, comparing their performance, and analyzing the effects of different parameters and dataset variations on training results.

Installation

Clone the repository and install the required packages using the provided environment.yml file:

git clone
cd rl_hw3
conda env create -f environment.yml
conda activate rl_hw3

Running the Code

python StandardQL_Training.py --dataset 50 # Train a standard Q-Learning model using a dataset of 50 episodes python BC_Training.py --dataset 50 # Train a Behavior Cloning model using a dataset of 50 episodes python CQL_Training.py --dataset 50 # Train a Conservative Q-Learning model using a dataset of 50 episodes

You may also specify the number of epochs, batch size, and other hyperparameters using command-line arguments. For more information, run the scripts with the --help flag such as 50, 150, 250, and 350 argument parameters supported.