Evaluating/testing the trained model about sevennet HOT 2 CLOSED

Hello, sorry for the delayed response,

1. We are currently in the process of introducing the feature you mentioned, the 'test_set' option. For now, you have the option to write your own python code using SEVENNet as package. However, please note that there isn't detailed documentation available for the classes yet. Alternatively, you can use LAMMPS with the desired model and then postprocess on the output.

2. This is my fault. The 'checkpoint' key in the yaml expects file path to the checkpoint you wish yo use for restarting. The correct input would be look like
   checkpoint: 'path_to_previous_run/checkpoint_best.pth'
   Note that if training continue on the same directory, it will overwrite log and .pt or .pth files. If this is not you want, please backup the previous training directory.

I'll mark this issue as resolved once the 'test_set' feature is updated. Thank you for the feedback.

Thank you.

from sevennet.

We finally updated our main branch with new feature 'sevenn_inference' to obtain correlation between prediction & reference.

After reinstalltion from main branch, you can get .csv of prediction & reference results by

sevenn_inference {checkpoint} {OUTCARs/POSCARs or .sevenn_data}

Sorry for the delated update. Since I re-wrote tons of code, you need to retrain the potential. You can check README.md for details new features.

Thank you.

from sevennet.