search

txie-93 / cgcnn

Crystal graph convolutional neural networks for predicting material properties.
MIT License
630 stars 305 forks source link

Data construction specifics #29

Open MichaelHopwood opened 3 years ago

MichaelHopwood commented 3 years ago

Hello, Thanks for your awesome work! I was wondering if you could provide more details regarding the data construction with the materials project API. Specifically,

  1. Which keys were queried to generate the features (Appendix Table 2)? a. Are group & period number included in the API or did you use other methods? b. Which keys in materialsproject held electronegativity, valence electrons (nsites?), etc.?

  2. What IDs are saved in atom_init? Do these IDs have to match with information in the CIF files?

Cheers!

txie-93 commented 3 years ago

Thank you for your question. First, the features are for elements, not for crystals. In atom_init, the keys are atomic number of the element, and the values are the initialization vectors for each element.

To generate these features, I used a package called mendeleev. For example, to get the group number for H, you can simply do,

from mendeleev import element

element('H').group_id
cywbreeze commented 10 months ago

The values, the initialization vectors are all the same, so they're all generic