Basic Behavioural Cloning and DAgger Implementation in PyTorch
- Define your policy network model in
model.py
. - Get appropriate states from environment. Here I am creating random episodes during training.
- Extract the expert action here from a
.txt
file or a pickle file or some function of states. - Run
python imitation_algos.py
.
- numpy
- pytorch
- tensorboardX
- tqdm
- Implement DAgger
- Make it compatible with OpenAI gym environments
- Add supporting algorithms to train further with Reinforcement Learning
- Comments in the code :P
tensorboard --logdir=runs --host localhost --port 8088