Develop with MEIDNet¶
meidnet is the reference implementation of the method: one shared latent space for crystal structures and their properties, a prototype-family design space checked by chemistry rules, and a search in the latent space for candidates with the properties you want. Everything on this site runs on it.
pip install meidnet # Python 3.10+; CPU is enough
meidnet demo # the published Perov-5 model, three oxide perovskites in a few minutes
meidnet studio # the same workflow in your browser
Where to start¶
| I want to | Page |
|---|---|
| see the whole workflow once | 5-minute quickstart |
| know whether my data fits | What data do I need?, bring your own dataset |
| train and design without installing anything | Run in Colab, the Studio |
| change what the search looks for | targets, properties, elements, family, rules |
| check candidates with a machine-learned potential | Screen stability with MACE |
| score my own model's structures | Benchmark compatibility |
| call it from Python | Python API, the configuration file, family files |
| read the reports | Reading the reports, interpreting a candidate |
The package¶
- Source: github.com/ABnano/MEIDNet (MIT). Issues and pull requests are welcome; getting help says what to include.
- Releases: PyPI and GitHub releases. The Hugging Face model page holds the published checkpoints.
- Command line: every command:
init,check,train,generate,studio,space,demo,screen,score,info,families,schema,download-data. - Extending it: a new material family is a YAML file (family files); a new rule is a Python function registered as a plugin (add a rule); a new property is a column (add a property).
Build on it¶
MEIDNet Matter is the first application built on the package: a FastAPI service and a React interface around meidnet.generate.Designer, with the model behind one backend interface so that other engines can be plugged in. Its source is at github.com/ABnano/MEIDNet-Matter; the ecosystem page says how the two relate.