Video Link: https://youtu.be/bZi8P6RNagU
Created agent to play games using different reinforcement learning techniques
Before executing the code, the following Python libraries need to be installed with a Python version of 3.7:
- mlagents
- pytorch
- tensorflow
- numpy
- wandb
- ipdb
For issues with tensorflow, activate a virtual environment through conda and install the following libraries:
- mlagents
- wandb
- ipdb
To train :
-
single agent using DQN algorithm, change the working directory to /submission/DQN and execute the Python script dqn_exp_unity_v3.py.
-
16 agents using DQN algorithm, change the working directory to /submission/MultiagentDQN and execute the Python script dqn_exp_unity_multiagent_v2.py.
-
Agent with PPO algorithm, change the working directory to /submission/PPO and execute the Python script ppo_unity.py.
After executing the script, when prompted, follow the steps displayed on the console to create a profile on wandb. We are using wandb to produce the results in a graph format continuously. The results can be seen through the wandb URL displayed after executing the command.