Skip to content

Bring your own dataset

This page takes a table of structures and properties to designed candidates. Every step writes a report that explains what happened; nothing requires Python.

0. Install

pip install meidnet

1. Describe your data: meidnet init

meidnet init --name my_oxides --table materials.csv \
             --id-column material_id --cif-column cif \
             --properties band_gap dielectric \
             --family perovskite_abx3 --variant oxide

This writes a commented meidnet.yaml. Open it: every line says what it does. If your CIFs are files instead of a column, set cif_column: null and structures_dir: structures/.

data:
  table: materials.csv
  id_column: material_id
  cif_column: cif
  properties:
    - {column: band_gap,   unit: eV}
    - {column: dielectric, unit: ""}
  val_fraction: 0.1
  max_sites: 20
  align_to_prototype: true

2. Is it usable? meidnet check

meidnet check meidnet.yaml

412 of 430 materials usable.
  skipped     14  does not match the perovskite_abx3 prototype   e.g. m_031, m_118, m_204
  skipped      4  missing property value                          e.g. m_077, m_300
report: runs/my_oxides/check_report.html

The report shows the verdict, the distribution of each property, the elements present and, for each skip reason, how to fix it (for example raise data.prototype_tolerance for distorted cells).

3. Learn the latent space: meidnet train

meidnet train meidnet.yaml            # 200 epochs by default
meidnet train meidnet.yaml --epochs 5 # a quick smoke test first

training_report.html answers three questions in plain words: how accurate is each property prediction (error vs natural spread → good / fair / weak), did the two modalities align (retrieval accuracy), and what were the training curves. The model is saved as runs/my_oxides/model.pt with the property names, units and normalisation inside; meidnet info runs/my_oxides/model.pt describes it.

Time

About 21 s per epoch per 11k structures on a laptop GPU; roughly 10× slower on CPU. Use Colab's free GPU (notebook 02) if you have none.

4. Design: meidnet generate

Set the targets in meidnet.yaml:

generation:
  family: perovskite_abx3
  variant: oxide
  objectives:
    - {property: band_gap,   loss: l2,       weight: 10000}
    - {property: dielectric, loss: at_least, weight: 5000}
  targets:
    - {band_gap: 2.0, dielectric: 20}
    - {band_gap: 3.0, dielectric: 20}
  per_target: 4
  exclude_elements: [Pb, Cd]

meidnet generate meidnet.yaml

generation_report.html shows, per target, the funnel (how many attempts each rule removed), one card per candidate (elements, cell, each rule's measured value and window, predicted vs target gauges, warnings about extrapolation) and examples of rejected compositions. CIFs and candidates.csv are in runs/my_oxides/generation/.

5. Explore live: meidnet studio

meidnet studio meidnet.yaml

Change a rule's limit, exclude an element or move a target and see immediately how many compositions pass and which are predicted closest; run searches and export the YAML. Studio guide.

6. Confirm

Predicted properties are estimates. meidnet screen runs/my_oxides/generation (MACE universal potential, optional extra) gives a first stability filter; DFT or experiment has the last word.