Conda env doesn't contain all needed packages?

[JSR694]

Hi there, When I try to run training for the single block-stacking task, i.e.:

python exps/train_op3/train_op3.py --variant [stack, pickplace, cloth] --debug 0

I get an import error for torchvision. But when I try to add it to the env (via conda install -n op3 torchvision), conda complains about conflicts (and spends forever trying to resolve them).

For context:

• Running this locally. Added "import op3.torch.pytorch_util as ptu; ptu.set_gpu_mode(True)" to the top of train_op3.py.
• Python couldn't find the op3 module, so I added it to my conda env (using conda-build).

Thanks!

[JSR694]

Looks like I needed to add torchvision=0.2.1 as a dependency in the conda env specification file.

Other problems with dependencies I noticed:

• lockfile-0.12.2 --> lockfile==0.12.2
• gym[all]==0.10.5lo --> gym[all]==0.10.5, but actually gym[all]==0.10.8 because otherwise I run into this error.
• (I had to add CUDA 10 as a dependency.)

Also needed to add OP3 itself as a dependency using conda develop ./OP3/; otherwise the scripts can't find the op3 module.

Hope this helps.