How to design id_props.csv?

[Anny-tech]

Hi I have a dataset with parent structures and their .cif files. However, I am testing the presence of dopants on specific sites characterized by specific point groups. I am especially interested in using the inverse model. My questions are:

1. Can I use .cif formats instead of POSCAR formats?
2. Apart from the filenames in the first column of the 'id_props.csv' file can I represent the above extra additions to the parent structure as a property column? I have 3 target value columns already.
3. So basically what I have in hand is, 1 column of filenames, 5 more columns, first being dopant element name, 2nd column~ Point Group name, 3rd- 5th columns ~ Target properties (numerical values)

I would be grateful if I can get some insight asap!

Thanks and Regards
Ankita

[knc6]

Hi @Anny-tech

1. If you are training inverse models, you can use file format as cif, see here.
2. Adding extra info is not supported inherently, but you can customize the make_alpaca_json function here to add your inputs.

Let me know if you encounter further issues, happy to help.