khrylx / transform2act Goto Github PK

When training gap task, there is an error showing that file assets/mujoco_textures/gap_texture.png is missing.

hello,I'm having some troubles when trying to run this project.
the following is my hardware version and my configured environment.

NVIDIA Geforce RTX 3080Ti, and because it's a 30 series graphics card, the minimum CUDA can be used is CUDA11.

python=3.7.0,torch=1.8.0+cu111, torch.cuda is available.

torch-geometric=2.1.0 with correct CUDA and PyTorch versions and Mujoco_py=2.1 can successfully be imported.

when run python design_opt/train.py --cfg hopper --gpu 0

The program does not report errors, but also no response

log file is empty

the tensorboard interface also has no data.

Additionally，I have successfully executed it in the PYTORCH=1.8.0+CU102 version on another server before.
Is this project unable to use CUDA11.1 for training?

1. There is no create_dirs argument in Config class's init function.

   Transform2Act/design_opt/train.py

   Line 22 in 9c49037

   cfg = Config(args.cfg, args.tmp, create_dirs=not (args.render or args.epoch != '0'))

2. When num_threads >= 2, the code gets stuck in the following line

   Transform2Act/design_opt/models/jsmlp.py

   Line 24 in 9c49037

   out[x_ind] = torch.addmm(b, x[x_ind], W.t())

   multiprocessing might be the reason. When num_threads=1 and Queue was not use, everything goes fine. I setup a conda environment as the instructions in README. Have you ever encountered this bug?

python design_opt/eval.py --cfg swimmer --save_video

loading model from checkpoint: results/swimmer/models/best.p
Traceback (most recent call last):
File "/home/dzyao/shiyuexin/shiyan/transform2act/design_opt/eval.py", line 28, in
agent = Transform2ActAgent(cfg=cfg, dtype=dtype, device=device, seed=cfg.seed, num_threads=1, training=False, checkpoint=epoch)
File "/home/dzyao/shiyuexin/shiyan/transform2act/design_opt/agents/transform2act_agent.py", line 38, in init
self.load_checkpoint(checkpoint)
File "/home/dzyao/shiyuexin/shiyan/transform2act/design_opt/agents/transform2act_agent.py", line 162, in load_checkpoint
self.policy_net.load_state_dict(model_cp['policy_dict'])
File "/home/dzyao/anaconda3/envs/pytorch2/lib/python3.9/site-packages/torch/nn/modules/module.py", line 2041, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for Transform2ActPolicy:
Missing key(s) in state_dict: "skel_gnn.gconv_layers.0.lin_rel.weight", "skel_gnn.gconv_layers.0.lin_rel.bias", "skel_gnn.gconv_layers.0.lin_root.weight", "skel_gnn.gconv_layers.1.lin_rel.weight", "skel_gnn.gconv_layers.1.lin_rel.bias", "skel_gnn.gconv_layers.1.lin_root.weight", "skel_gnn.gconv_layers.2.lin_rel.weight", "skel_gnn.gconv_layers.2.lin_rel.bias", "skel_gnn.gconv_layers.2.lin_root.weight", "attr_gnn.gconv_layers.0.lin_rel.weight", "attr_gnn.gconv_layers.0.lin_rel.bias", "attr_gnn.gconv_layers.0.lin_root.weight", "attr_gnn.gconv_layers.1.lin_rel.weight", "attr_gnn.gconv_layers.1.lin_rel.bias", "attr_gnn.gconv_layers.1.lin_root.weight", "attr_gnn.gconv_layers.2.lin_rel.weight", "attr_gnn.gconv_layers.2.lin_rel.bias", "attr_gnn.gconv_layers.2.lin_root.weight", "control_gnn.gconv_layers.0.lin_rel.weight", "control_gnn.gconv_layers.0.lin_rel.bias", "control_gnn.gconv_layers.0.lin_root.weight", "control_gnn.gconv_layers.1.lin_rel.weight", "control_gnn.gconv_layers.1.lin_rel.bias", "control_gnn.gconv_layers.1.lin_root.weight", "control_gnn.gconv_layers.2.lin_rel.weight", "control_gnn.gconv_layers.2.lin_rel.bias", "control_gnn.gconv_layers.2.lin_root.weight".
Unexpected key(s) in state_dict: "skel_gnn.gconv_layers.0.lin_l.weight", "skel_gnn.gconv_layers.0.lin_l.bias", "skel_gnn.gconv_layers.0.lin_r.weight", "skel_gnn.gconv_layers.1.lin_l.weight", "skel_gnn.gconv_layers.1.lin_l.bias", "skel_gnn.gconv_layers.1.lin_r.weight", "skel_gnn.gconv_layers.2.lin_l.weight", "skel_gnn.gconv_layers.2.lin_l.bias", "skel_gnn.gconv_layers.2.lin_r.weight", "attr_gnn.gconv_layers.0.lin_l.weight", "attr_gnn.gconv_layers.0.lin_l.bias", "attr_gnn.gconv_layers.0.lin_r.weight", "attr_gnn.gconv_layers.1.lin_l.weight", "attr_gnn.gconv_layers.1.lin_l.bias", "attr_gnn.gconv_layers.1.lin_r.weight", "attr_gnn.gconv_layers.2.lin_l.weight", "attr_gnn.gconv_layers.2.lin_l.bias", "attr_gnn.gconv_layers.2.lin_r.weight", "control_gnn.gconv_layers.0.lin_l.weight", "control_gnn.gconv_layers.0.lin_l.bias", "control_gnn.gconv_layers.0.lin_r.weight", "control_gnn.gconv_layers.1.lin_l.weight", "control_gnn.gconv_layers.1.lin_l.bias", "control_gnn.gconv_layers.1.lin_r.weight", "control_gnn.gconv_layers.2.lin_l.weight", "control_gnn.gconv_layers.2.lin_l.bias", "control_gnn.gconv_layers.2.lin_r.weight".