Model to be used for inference

[ksr-burn]

Hello,

There is no model available in the trained_models directory referenced by the run_model.py script. Is there a version of the trained model available to test the program?

[bafgreat]

Thank you very much for your feedback and interest in using fairmofsyncondition.

We intentionally did not distribute trained PyTorch models as serialized .pt files.
Saving and sharing raw .pt checkpoints is generally discouraged due to long-term compatibility,
versioning and reproducibility concerns. Instead, the model architecture and weights are encapsulated
directly in code.

Model access

The trained model parameters are found MetalSaltGNN_Ablation class , which can be imported from fairmofsyncondition.call_model.utils_model. You can effectively rebuild the model by calling this

This is exactly what is happening in fairmofsyncondition.call_model.utils_model.get_models function.

Make prediction

The easiest way to make prediction is by running the following line of codes:

from fairmofsyncondition.call_model.run_model import Synconmodel

# Call the prediction pipeline 
ml_data =  Synconmodel(input)

# if you only one the prediction 
prediction = ml_data.predict_condition()

# If you want your prediction to be written to a file
# simply run the following 
ml_data.outfile = 'my_prediction.txt'
ml_data.compile_data()

Note

The input argument in ** Synconmodel(input)** can:
1. An ASE Atoms object
2. Any ASE readable structure file (e.g. .cif)
3. A PyTorch Geometric Data object, created using:
data = fairmofsyncondition.read_write.coords_library.ase_to_pytorch_geometric(input)
where input can be (1 and 2)

Do let me know if this answers your question and whether you have any other worry.

[bafgreat]

Furthermore, if you wish to call the model from the command line, simply run the following.

 fairmofsyncondition_syncon my_mof.cif -o my_mof_report.txt

[ksr-burn]

The fairmofsyncondition.call_model.utils_model function loads the model from the checkpoint path which directs it to the trained_models folder where the trained_models/Metal_salts_{config_name}_tmp_test.pt file is missing

[bafgreat]

Oh you are a legend. I actually had it in my .gitignore thus it worked on the pip installable version and not the git version. I have fixed that now. Please git pull and it should be fine now. Please do let me know if you into any issue.