search

kilianhae / discrete_DPPM_Graphs

Diffusion Models for Graphs Benefit From Discrete State Spaces
26 stars 0 forks source link

Question about running the code directly #1

Open Frankie123421 opened 2 years ago

Frankie123421 commented 2 years ago

Hi, thanks for sharing the code. I tried to run the code directly without slurm, and the command is like "python ppgn_simple.py -c ./config/train_ego_small_ddpm_16.yaml". However, I found that some args using in ppgn_simple.py is missed in the config file, for example, config.noiselevels. Could you please provide some guidance to fix it? I don't know if it's because the remaining setting is done in files related to gridsearch. Thanks in advance.

kilianhae commented 2 years ago

Hi, thanks for taking the time to look at my project. Yes some parameters are set in the gridsearch.py scripts but I have added complete configuration files now. So running it as you mentioned should be possible: "python ppgn_simple.py -c ./config/train_ego_small_ddpm.yaml". Also I will update the gridsearch files to be able to run them without slurm tomorrow.

kilianhae commented 2 years ago

Also, make sure to setup your wandb details in the train_main functions or comment the corresponding lines out, I have added this into the README as well.

Frankie123421 commented 2 years ago

Thanks for your reply and kind reminder. I will retry it.

Frankie123421 commented 2 years ago

Hi, I wonder if you ran the experiment without gpu, as I can't seem to find any about "cuda" in the code.

kilianhae commented 2 years ago

Hi, yes everything is run as cuda as long as cuda is available. The config object gets processed in the process_config() function in /utils/arg_helper.py. There the lines: dev=torch.device(“cuda”) if torch.cuda.is_available() else torch.device(“cpu”) config.dev = dev append the .dev attribute to the config object, which is called during training and sampling to specify the device to use.

kilianhae commented 2 years ago

BTW I have tested it now with running it directly as a python shell. There were some implementations that were still not working without slurs. This issue should now be fixed as I have tested it. Please let me know if anything is not working for you.