DeepFM for MOFs (DeepFM4MOF)

This code provides implementation of DeepFM for prediction of gas adsorption properties of MOFs, and the results of the paper (DeepFM4MOF) can be reproduced using this code.

├── utils.py   
│   ├── create_hmof_dataset  # date processing, devide the data based on arguements
├── layer.py  
│   ├── FM_layer    # FM component
│   ├── Dense_layer # Deep component
├── model.py  
│   ├── DeepFM      # DeepFM model
├── train.py 
│   ├── main        # train and show the results

---

If used to impute missing values in the material-property matrix:

python train.py --coldstart=false

If used to predict target properties using cold-start experiments:

python train.py --coldstart=true --property=1 --targetfrac=0.9

(--property (int, default=1) is the predicted property, range of it is 1 to 28, --targetfrac (float, default=0.9) is the fraction of the test set of the target property, say it's 0.9, then 90% of the target property is separated from the dataset, it will be uesd to test the trained model,range of it is 0 to 1)

If you used this code in your paper, please cite Finding the optimal CO2 adsorption material: Prediction of multi-properties of metal-organic frameworks (MOFs) based on DeepFM.Thank you.