Descriptor notebook for AI Toolkit
Here is the content we got from @jpd47 https://datashare.mpcdf.mpg.de/s/mdX3geCMMMdYySn
Some more text about it can be found in the second last section here https://docs.google.com/document/d/1d3FdO_7vY6l1xGEC1tIWhotCpg4zGX10mdQXRZkSZqo/edit?usp=sharing
This is @jpd47 's input:
Here’s a first attempt at the notebook we discussed. It produces three different maps, two using a local data file (included in the tarball) and one from a direct nomad query.
The tarball contains
- the notebook
- local data file
- requirements.txt (tested with a fresh venv)
- Cached files containing precomputed descriptors - stored as numpy arrays without pickle. This is less important for SOAP but quite useful for MACE.
Let me know if you have any suggestions/trouble + here are some screenshots of what the final maps should (roughly) look like. The KPCA is deterministic but the UMAP is not.
This is what it supposed to look like: