Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Encountering difficulties in reproducing the results of the paper. #9

Open
Cristallgalaxy opened this issue Sep 28, 2024 · 2 comments

Comments

@Cristallgalaxy
Copy link

image
I have run this code using the single-chain CG polymer dataset and encountered some issues.

  1. It seems that the default hyperparameters in graphwm/conf/model/chain.yaml do not use PnR for the single-chain CG polymer dataset. I have found that if I want to use PnR for this dataset, I should add score_gn_hparams: {'latent_dim': 128, 'units': 128, 'layers': 2, 'mp_steps': 7}, sigma_begin, sigma_end, sigma_level, and anneal_power in graphwm/conf/model/chain.yaml, but what should I set these parameters to? Does PnR perform better than GNS in the single-chain CG polymer dataset?
  2. I have run this code using the GNS model and the single-chain CG polymer dataset with the default hyperparameters in train.yaml and eval.yaml. I then got a .pt file with the keys: ['cg_u_pos', 'cg_weights', 'rollout_u_pos', 'rollout_prop', 'bonds', 'n_bond', 'position', 'rgs', 'target', 'target_rgs', 'particle_types', 'n_particle', 'keypoint', 'n_keypoint', 'cluster', 'cg_bonds', 'n_cg_bond', 'time_elapsed', 'model_params', 'eval_cfg']. 'rollout_prop' is what I need; it stores the rgs of each step. I saved this data in a .txt file. The true rgs is stored in a .h5 file in polymer_test_5M. I can calculate the MAE with the predicted results and true results, but I find it difficult to achieve the MAE result of 1.4 mentioned in the paper. I think I might have made a mistake somewhere; how should I calculate the MAE?

Any suggestions would be very helpful to me; thank you very much.

@kyonofx
Copy link
Owner

kyonofx commented Oct 1, 2024

Hi,

  1. As described in the paper, GNS was used for the single-chain CG polymer dataset. PnR would be too slow for the timestep setting.
  2. Have you tried the pretrained checkpoint available in this repo? Note that the MAE of 1.4 is from running the simulation for 5M tau then compute the mean radius of gyration over these trajectories.

@Cristallgalaxy
Copy link
Author

I'm sorry, but how do I use the pretrained checkpoint? When I try to use it to train the model, I encounter an error: FileNotFoundError: [Errno 2] No such file or directory: '/data/tml/code/xiangfu/mlcgmd_models/chain_gns_smalllr/epoch=1-step=48040.ckpt'. Should I modify a certain part of the code?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants