search

ARISE-Initiative / robosuite-benchmark

Benchmarking Repository for robosuite + SAC
53 stars 15 forks source link

TypeError: no default __reduce__ due to non-trivial __cinit__ #2

Closed harishbommakanti closed 3 years ago

harishbommakanti commented 3 years ago

While trying to run an experiment using python scripts/train.py, the program runs into an error while trying to save the data. This seems to happen when I specify specific .json files to run too. Attached is the run log.

I'd appreciate any help. Thanks! rb_bench_errormessage.txt

cremebrule commented 3 years ago

Huge thanks for bringing up this issue -- there turned out to a be a small (but important) local change on my machine that wasn't added to this repo. Can you please uninstall the current conda environment (rb_bench) and re-install using the updated README?

harishbommakanti commented 3 years ago

I removed the environment, started from a clean slate, and I got the error which https://github.com/ARISE-Initiative/robosuite/issues/111 is talking about. I remember that I got this maybe a week ago but the problem went away after https://github.com/ARISE-Initiative/robosuite/pull/112 if I'm remebering correctly, I'm not sure why I'm getting the error again. Maybe the wrong robosuite version in the requirements.txt/the yaml needs to be updated, or its just a problem with my config.

Here's the log message for reference, but it's just talking about the missing robosuite/controllers/interpolators module: rb_bench_errormessage.txt

cremebrule commented 3 years ago

Good catch -- yes, this is a robosuite version issue. Can you run pip install robosuite==1.0.1?

harishbommakanti commented 3 years ago

Yeah that looks like it works. It at least got thorough epoch 0! I'll run a full experiment later today and keep you posted.

cremebrule commented 3 years ago

Hi @harishbommakanti , how did the experiments turn out? If everything is working, please close the issue. :)

harishbommakanti commented 3 years ago

I left it to train over the weekend and the experiment finished all 2000 epochs! I ran the rollout and that worked too so I don't think I'm seeing any more errors.