How to make a multi-target regression with m3gnet model? about matgl

[kenko911]

Hi @KirillKulaev, thanks for reporting the error. I would like to clarify a bit on your regression problem. I think you are referring to the multiple values per target for each structure/graph but not another way around i.e. single value but multiple targets for each structure/graph. Regarding the multiple values per target, the framework of collate_fn needs to be modified so that it works for your purpose. I will push the fix soon. Thanks!

[KirillKulaev]

Hi @KirillKulaev, thanks for reporting the error. I would like to clarify a bit on your regression problem. I think you are referring to the multiple values per target for each structure/graph but not another way around i.e. single value but multiple targets for each structure/graph. Regarding the multiple values per target, the framework of collate_fn needs to be modified so that it works for your purpose. I will push the fix soon. Thanks!

Thank you very much for your answer!

[kenko911]

Hi @KirillKulaev, I just pushed the fix for the collate_fn and now it should work. Please pull the latest version of MatGL and take the unit test of model training (tests/utils/test_training.py) for multiple values per target with M3GNet as a reference to modify your script. I would like to stress that this is an experimental feature and I am not sure how good is the accuracy of the M3GNet model for your purpose.

[KirillKulaev]

Hi @KirillKulaev, I just pushed the fix for the collate_fn and now it should work. Please pull the latest version of MatGL and take the unit test of model training (tests/utils/test_training.py) for multiple values per target with M3GNet as a reference to modify your script. I would like to stress that this is an experimental feature and I am not sure how good is the accuracy of the M3GNet model for your purpose.

Thank you very much, everything works well