I have a brief question about the results that I am supposed to get on wandb as well as from running main.py and eval_policy.py.
I'm attempting to train an ordered push policy. I run main.py -c config/ordered_push. In the respective Config.yaml file I specify latentRepPath: 'latent_rep_chkpts/dlp_push_5C' for pre-trained DLP and train with numObjects = 2. I run policy_eval.py config/ordered_push with the respective checkpoint found in model_chkpts and the same number of objects.
I am unsure of what I am supposed to expect in the results. My issues are as follows:
There are several .gif files that I obtained in wanbd: Eval Goal Image, Goal Image - Frontview, Goal Image - Sideview, Episode Video - Frontview, Episode Video - Sideview, and Eval Episode Video. I'm getting videos for both main.py and policy_eval.py. I'm not sure about what these videos represent and at which point in the training they are recorded.
I notice that the code runs after 800 episodes. I'm probably misunderstanding the code a little bit, but I'm currently expecting that the code should stop as it is specified in the config file.
I would be curious to see what are my expected graphs for obs_rms_std and obs_rms_mean should look like. I see the error decrease, but only from 0.68 to 0.655 and -0.03 to -0.04 respectively (which tells me that I'm probably not learning much). Would you mind telling me what results you got on that end?
Thank you very much for your help!
Hanna
p.s. Would it be possible to email you for further personal issues I have?
While trying to run your code, I ran into several dependency issues.
In the prerequisites section, you mention having torch 2.0.1 and stable-baseline3 1.5.0. However, your requirements.txt has torch 2.1.2. Furthermore, I can't install gym 0.21.0 without downgrading to python 3.7. I have more similar issues that require newer versions of stable-baseline3.
Also, I have permission denied when trying to run python main.py -c config/push_t (for wandb).
Would you mind helping me to get started running your code in conda?