How to view the training model and how to visualize it

[hongjianjiang]

How to view the training model and how to visualize it. How to use the checkpoint on the result folder

[keiohta]

Hi @swordJJ , thanks for using tf2rl.

If you run a script, logs and models are saved under results/%Y%m%dT%H%M%S.%f. For example, following is an example of training a SAC agent on Pendulum-v0 environment.

$ cd tf2rl_root

# train SAC agent on Pendulum-v0 env.
$ python examples/run_sac.py --env-name Pendulum-v0
...
17:38:32.824 [INFO] (trainer.py:122) Total Epi:    77 Steps:   15400 Episode Steps:   200 Return: -3.3546 FPS: 235.26
17:38:33.697 [INFO] (trainer.py:122) Total Epi:    78 Steps:   15600 Episode Steps:   200 Return: -0.8957 FPS: 229.31
...

# you can find logs and saved models under `results/%Y%m%dT%H%M%S.%f`
$ ls results/20211027T174421.731912_SAC_/
20211027T174421.890139.log  ckpt-2.data-00000-of-00001  ckpt-4.index                                         git-head.txt
args.txt                    ckpt-2.index                command.txt                                          git-log.txt
checkpoint                  ckpt-3.data-00000-of-00001  environ.txt                                          git-status.txt
ckpt-1.data-00000-of-00001  ckpt-3.index                events.out.tfevents.1635324261.ohta-SZ.11377.491.v2
ckpt-1.index                ckpt-4.data-00000-of-00001  git-diff.txt

# rollout with the trained model by adding `--evaluate` and specify `--model-dir`. you can also visualize the rollouts by adding `--show-test-progress`
$ python examples/run_sac.py --env-name Pendulum-v0 --evaluate --model-dir results/20211027T174421.731912_SAC_/ --show-test-progress

Does this answer the question?

[hongjianjiang]

Thanks for your reply very much. Due to the limit knowledge about it, it bothered me for a long time. Thank you.