Simple pytorch implementation of MADDPG for an cooperative task with a shared actor and critic network.
Run
python train.py
while being in the src folder. The parameters can be changed by either creating a new config file and then
python train.py --cfg <config-file path>
or directly i.e.
python train.py --lr_critic 0.01
which takes the parameters from the default file but ovverrides the given parameter/s.
To plot the performance of the agent, run
python plot.py --run_folder <path to the run-folder>
to plot the losses, Q-values and performance of the agent over time.
To test the performance of the agent, run
python test.py --run_folder <path to the run-folder>
to test and render the performance of the newest saved agent weights.
