Skip to content

Latest commit

 

History

History
50 lines (35 loc) · 2.58 KB

README.md

File metadata and controls

50 lines (35 loc) · 2.58 KB
Raw

re2fractive

The aim of this project is to design and discover materials with high refractive indices by exploiting new and existing databases accessible through OPTIMADE, machine learning predictions (via MODNet) and high-throughtput DFT calculations (via atomate2 and jobflow-remote, all within a dynamic active learning framework.

This repository contains a Python package, re2fractive that implements some of this functionality, with the aim to grow it to a generic package for other properties.

The resulting VT-2k dataset is available on the Materials Cloud Archive:

V. Trinquet et al., Optical materials discovery and design with federated databases and machine learning, Materials Cloud Archive 2024.114 (2024), DOI: 10.24435/materialscloud:5p-vq

with its own OPTIMADE API at https://optimade.materialscloud.org/archive/5p-vq/.

This repository accompanies the Faraday Discussions paper:

V. Trinquet, M. L. Evans, C. Hargreaves, P-P. De Breuck, G-M. Rignanese, "Optical materials discovery and design with federated databases and machine learning", Faraday Discussions (2024) DOI: 10.1039/D4FD00092G (preprint: 10.48550/arXiv.2405.11393).

The active learning campaign described there can be repeated (modulo some details around atomate2 & jobflow-remote setup) with:

from re2fractive.campaign import Campaign, LearningStrategy
from re2fractive.datasets import NaccaratoDataset, MP2023Dataset, Alexandria2024Dataset
from re2fractive.oracles import Re2DielectricMaker

learning_strategy = LearningStrategy(
    max_n_features=100,
    feature_select_strategy="always",
    hyperopt_strategy="always",
)

campaign = Campaign.new_campaign_from_dataset(
    NaccaratoDataset,
    datasets=[MP2023Dataset, Alexandria2024Dataset],
    oracles=[("refractive_index", Re2DielectricMaker)]
    learning_strategy=learning_strategy
)

campaign.run(epochs=8)

Some functionality is still missing from the first public release:

  • Direct integration with atomate2/jobflow-remote workflows for automatic job submission after candidate selection.
  • Automatic selection according to custom acquisition functions.
  • Automatic addition of new data (via a new learning strategy option) that reruns OPTIMADE queries with a date range and screens new entries/databases.