Bugfix/invalid structures

[crhysc]

Encountered issues where an invalid lattice prediction will crash the evaluate() loop and crash the train/test script. Here is a sample error that I got on test structure #44:

genmat None
Traceback (most recent call last):
File "/users/crc00042/repositories/atomgpt/atomgpt/inverse_models/inverse_models.py", line 654, in
main(config_file=args.config_name)
File "/users/crc00042/repositories/atomgpt/atomgpt/inverse_models/inverse_models.py", line 639, in main
evaluate(
File "/users/crc00042/repositories/atomgpt/atomgpt/inverse_models/inverse_models.py", line 241, in evaluate
+ Poscar(gen_mat).to_string().replace("\n", "\n")
File "/users/crc00042/miniforge3/envs/ramangpt/lib/python3.10/site-packages/jarvis/io/vasp/inputs.py", line 56, in to_string
+ str(self.atoms.lattice_mat[0][0])
AttributeError: 'NoneType' object has no attribute 'lattice_mat'

These code changes will prevent evaluate() crashes and make two benchmark CSVs. One will contain structures that have valid reconstructions, and the other will contain structures whose reconstructions are invalid. Let me know what you think

[knc6]

Thanks @crhysc . Did you check if the pytests/examples on the README.md file pass with the new PR?

[crhysc]

I patched inverse_models.py and inverse_predict.py, and the REAMDME tests pass. Should be good to go