volcorr models
Trained ensembles for volcorr: prediction of y = ln(V_exp / V_ref), the volume error of a PBE/PBE+U relaxation with Materials Project settings (tiers A to D) or of a MACE-MPA-0 relaxation (tier E), relative to experiment.
| file prefix | input | model |
|---|---|---|
| A_formula | formula | 5 × LightGBM (L1) |
| B_spacegroup | formula + space group | 5 × LightGBM (L1) |
| C_cif | experimental structure | 5 × LightGBM (L1) |
| D_vdft | formula + space group + PBE(+U) volume | 5 × LightGBM (L1) |
| E_mace | MACE-MPA-0 relaxed structure | 5 × (LightGBM + MLP, median-regression stack) |
| F_finished | finished PBE(+U) relaxation: relaxed structure, band gap, magnetization | 5 × LightGBM (L1) |
| G_finished_mp | as F_finished, plus formation energy and energy above hull | 5 × LightGBM (L1) |
registry.json lists features, SHA-256 hashes, stack weights and held-out residual quantiles.
LightGBM files are gzipped text boosters; MLP files are numpy arrays evaluated by
volcorr.model.mlp_predict. Trained on 40,629 matched Materials Project and ICSD entries whose PBE(+U) volume is within
10 % of experiment.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support