Unsupervised structural identification via a multilayer Self-Organizing Map (SOM): clusters atoms into a per-atom cluster_id from features produced by selected feature plugins (e.g. coordination-analysis, centrosymmetry-parameter, steinhardt-order-parameters).
vpm install @voltlabs/multisommultisom <output_base> --layers '<json>' [options]| Argument | Required | Default | Description |
|---|---|---|---|
<output_base> |
yes | — | Working/output directory; reads upstream feature parquets here and writes outputs as {output_base}_<suffix>. |
--layers |
yes | — | JSON array of SOM layers. Each layer: label, features, scaling (normal/standard/robust), f, sigma, eta, N, batched, batch_size. |
--training |
no | reference |
reference (stable cluster IDs across frames) or per_frame (retrain each frame). |
--mapping |
no | linear |
Cluster ID mapping: linear, godel, or default. |
--seed |
no | 1982 |
Random seed. |
--frameIndex |
no | 0 |
Index of the current frame in the filtered trajectory. |
| Output file | Exposure | Exporter → artifact |
|---|---|---|
{output_base}_atoms.parquet |
MultiSOM Model | AtomisticExporter → glb |
{output_base}_multisom.parquet |
MultiSOM | — (listing-only) |
Full input contract and examples: https://docs.voltcloud.dev/docs/plugins