amkrajewski/MPDD-X

[ENH]

Closed this issue · 5 comments

cwf7144 commented

We are happy to quickly add any structure-informed model which is a plug-and-play addition to our infrastructure, which will generally mean it either:

  • Utilizes already implemented representation, feature set, or its subset. Please check below as appropriate:

    • My model uses KS2022 or its subset.
    • My model uses ALIGNN graph representation.
    • My model is a refitting or a variation of CHGNet
    • My model is available from a public citable source like Zenodo (preferred) or figshare.
    • My model can be fetched within 30s, corresponding to ~2GB on Zenodo or 250MB on figshare.
    • My model can run single-threaded in under 500ms / structure
    • My model can run single-threaded in under 5s / structure

  • Is a lightweight tool, which can be quickly installed and run. Please check below as appropriate:

    • My model is well maintained and has been used in past research.
    • My model is actively tested at least weekly and has been well-maintained for at least 6 months.
    • My model is available from a public citable source like Zenodo (preferred) or figshare.
    • My model can be fetched within 30s, corresponding to ~2GB on Zenodo or 250MB on figshare.
    • My model can run single-threaded in under 500ms / structure
    • My model can run single-threaded in under 5s / structure

If it does not fall into the above categories, please dont get discouraged and still let us know! It will be much more work on our side, but we will probably be happy to work with you. To get started on that, please explain it in a few sentences.

github-actions commented

Hello @cwf7144! Please comment below on this issue while attaching a ZIP file with atomic structures (POSCAR/CIFs) to get started! You can put keywords describing it in the body of the comment.

As soon as you do it, your atomic structures will be processed through several tools, described below, and return a neat Markdown report.

  • pySIPFENN framework, returning (1) array of descriptors (feature vectors) in Numpy .npy and CSV formats you can use for your ML modelling, alongside formation energy predictions.
  • ALIGNN framework, returning (1) results from 7 ALIGNN models specified here and (2) compressed graph representation files.
  • CHGNet model, returning (1) energy prediction for your input, (2) CHGNet-relaxed structures in the same format (POSCAR/CIF) as your input, and (3) energy prediction for the relaxed structures.
github-actions commented

👍 [1/3] I found the ZIP file linked in your comment. I will now attempt to download it and validate contents!

github-actions commented

👍 [1/3] I found the ZIP file linked in your comment. I will now attempt to download it and validate contents!