Page 1 of 1

How to continue training a model with train_ppo ?

Posted: Tue Jul 24, 2018 8:05 pm
by underactuated
I trained a model in the pybullet_ant environment with

python -m pybullet_envs.agents.train_ppo --config=pybullet_ant --logdir=ant

The trained model is now stored in ant/xxxxx (where xxxxx is a time-stamped subdirectory). Can I force train_ppo to load the trained model in a new training session? So I would continue training it, instead of starting from scratch?

Thank you.

Re: How to continue training a model with train_ppo ?

Posted: Tue Nov 06, 2018 5:07 am
by Erwin Coumans
If you want to resume a previously started run, add the --timestamp=<time> flag to the last command and provide the timestamp in the directory name of your run.

See also the more new version of PPO for more info on the --timestamp option.
https://github.com/google-research/batch-ppo

(the new version may not be compatible with PyBullet though, may need some work)