search

jiaor17 / DiffCSP

[NeurIPS 2023] The implementation for the paper "Crystal Structure Prediction by Joint Equivariant Diffusion"
MIT License
59 stars 18 forks source link

some trouble in training #16

Open HSP23SCM26J opened 1 month ago

HSP23SCM26J commented 1 month ago

Hi JiaoR, Thank you for your nice work! During the preparation of training, I have a trouble such this: InstantiationException('Error in call to target \'diffcsp.pl_data.dataset.CrystDataset\':\nTypeError("CrystDataset.init() missing 4 required positional arguments: \'save_path\', \'tolerance\', \'use_space_group\', and \'use_pos_index\'")\nfull_key: datasets.train')

I didn't find any information about the missing arguments in the yaml file, could you please give me some advice?

jiaor17 commented 1 month ago

Hi, Thanks for your interest! The parameter save_path determines the absolute path to save the pre-processed dataset. For example, in the mp-20 yaml file, the save_path of the training set is ${data.root_path}/train_ori.pt. And the other three parameters, tolerance, use_space_group, and use_pos_index, are previously used for some spacegroup-related explorations and not quite related to the current version of DiffCSP. You can set them as the following default settings to ignore them:

tolerance: 0.1
use_space_group: false
use_pos_index: false

Hope these could help you!

HSP23SCM26J commented 1 month ago

Thank you very much!

Actually about the implementation of the command given the specific dataset name (carbon_24), the generated yaml file automatically is different with the haparams yaml file in carbon_24 data file, there is not any value of key data