Implementation of 3 RL Algorithms

First, install box2d from source:

git clone https://github.com/jonasschneider/box2d-py/
cd box2d-py
python setup.py build
python setup.py install

Then, install visdom from source:

git clone https://github.com/facebookresearch/visdom
cd visdom
pip install -e .
easy_install .

Install TensorFlow and PyTorch following the instructions on the website.

Finally, check all required packages have been installed by

pip install -r requirements.txt

Train a cloned policy by

python imiatation.py --train --num_episodes [NUM_EPISODES] --log_dir [LOG_DIR]

where NUM_EPISODES is the number of expert episodes in the training set.

Run python imitation.py -h for more options.

First, start a visdom server by

python -m visdom.server

To train a REINFORCE agent with default parameters, run

python reinforce.py --task_name [TASK_NAME]

You can navigate to http://localhost:8097 to see the learning curve and other plots.

Run python reinforce.py -h to see the hyperparameters that can be set.

First, start a visdom server by

python -m visdom.server

To train an A2C agent with default parameters, run

python a2c.py --task_name [TASK_NAME] -n [N]

where N is the number bootstrapping steps.

You can navigate to http://localhost:8097 to see the learning curve and other plots.

Run python a2c.py -h to see the hyperparameters that can be set.

wentaoyuan/rl3