[Closes #36] Creates the soft-actor critic from a Udemy course. This course goes over one way to implement the original soft actor-critic algorithm. The references to (Haarnoja, 2018) reference this paper.
To test the agent, you should be run the python script (in a virtual environment) like so:
python3 train_agent.py --group udemy_sac_agent
This should automatically start training the agent and save the best model parameters every 1000 timesteps. You should be able to run the agent with pretrained weights like so:
[Closes #36] Creates the soft-actor critic from a Udemy course. This course goes over one way to implement the original soft actor-critic algorithm. The references to (Haarnoja, 2018) reference this paper.
To test the agent, you should be run the python script (in a virtual environment) like so:
This should automatically start training the agent and save the best model parameters every 1000 timesteps. You should be able to run the agent with pretrained weights like so:
Note that the
/path/to/model_filenameshould not include the.pth.tarextension.The hyperparameters are not yet tuned but after ~200k steps I was able to get a consistent score around 2700.