How can I train the model with my own data?

I am interested in your work and I really appreciate your efforts in retraining all the baseline models on the same datasets. I would like to know that would you publish your datasets as well as the codes? Thank you very much!

Hi, thanks for providing the nice codebase! I am currently trying to reproduce the results on MP e_form. However, I can only get a list of ckpts like the one below and the test MAE is around 0.91. The other tasks on MP didn't have a similar issue and the numbers look correct.

Do you have any thoughts of this?
Thank you!

Also I would like to ask if you can provide a brief guideline on how can we revise the code/config to reproduce the numbers as in the paper and where should we put the dataset files in. Thanks a bunch!

I get an exception at line 132 in matformer/train.py file when I try to run the code. The exception prints as follows - "Check 'model'
error in converting to training config!". The problem is the config dictionary cannot be converted to the TrainingConfig type object. Can you please take a look at it?

I am trying to reproduce the results in your paper. Unfortunately, with the learning rate and epochs in your paper, I can not reproduce the results for Bulk modulus and Shear modulus. Could you please specify how to reproduce the results?
I just followed your descriptions: download the data and put it under the folder "data", and ran the code with the script you provide in the repo.

Hi, @YKQ98! I asked the questions in another thread, but you might not see them. So, I open an issue here to ask them again. Sorry for bothering you!
I notice there is an argument 'Voronoi' for 'neighbor_strategy' but you didn't implement it. What kind of strategy does it refer to? Have you tried it by any chance and how does it perform? Thanks!

Hello, I would like to ask how the crystal structure of a cif file can predict the formation energy after I have trained a crystal structure and formation energy model and obtained .pt file. Can you write some code? Thank you very much!

Hi @YKQ98 !

Thanks for this nice repo! I was training on MP e_form but I kept getting -1 for train MAE. Do you have any suggestions/explanations about this? Looking forward to your reply.

For tasks of Bulk Moduli and Shear Moduli, we follow GATGNN, the recent state-of-the-art method for these two tasks, and use the same training, validation, and test sets, including 4664, 393, and 393 crystals.

GATGNN is NOT true state of the art algo for this
You may want to try with the latest model
https://github.com/usccolumbia/deeperGATGNN

In the Training and Prediction section, there is a command called "python setup.py". But it needs an additional argument. Also, there is no train.py folder in both the dataset directory.

How to reproduce the result for shear modulus for the Material project dataset, when I run this code train_mp.py by changing prop[0] to prop[-1] and epochs to 300 I got 16.347 MAE. I also convert the GPa to log(GPa), also I got the same type of MAE score.

Hi!
I was trying to run an exp on band gap prediction. But I got some strange numbers that differ a lot from what you got in the paper.
Do you have any thoughts on this? Thanks!