Development of a GridWord environment to train a DQN (Deep Q-Network), which was also developed as part of the project.
Deep Q-Learning is an extended version of the traditional Q-Learning algorithm, which uses deep neural networks to estimate Q-values. Classic Q-Learning is effective in environments with a limited and fixed number of states, but it encounters problems with extensive or continuous state spaces due to the size of the Q-table. Deep Q-Learning overcomes this limitation by replacing the Q-table with a neural network capable of estimating Q-values for each state-action pair.
To visualise DQN learning on the Gridword, you can run this project.
git clone python -m venv venv
source ./venv/bin/activate pip install -r requirements.txt python train_dqn.pyTranslated with DeepL.com (free version)