← All docs

Classification

Classification turns your curated data into a model that labels particles automatically.

From data to a model

The path is a pipeline of steps:

  1. Compose a from your collections — the class space the model will learn.
  2. Export a — a frozen, versioned snapshot with train/validation/test splits.
  3. Train a on that dataset.
  4. Evaluate it — confusion matrix, misclassifications, metrics.
  5. Deploy it so it predicts on incoming particles.

Pipelines and nodes

A arranges

s into a cascade (L1 → L2 → …): particles are sorted coarsely first, then refined. Every node is itself a classifier — there's no special "non-classifier" node. The compendium defines *which* classes participate; the tree defines *how* they're arranged.

Basic vs Advanced

Train a Model vs Compose a Pipeline. Train a Model bootstraps a single classifier node end to end. Compose a Pipeline wires together models you’ve already trained. The Advanced builder lets you define and auto-train a whole cascade at once.

Predictions and labels

A deployed model produces predictions, which feed each particle’s

. Your [annotations](/docs/annotations) always take precedence over a prediction.

Quick start

  1. Assemble a (see Collections).
  2. Export a and train a model.
  3. Review the evaluation, then deploy.