Training custom classifiers

This tutorial outlines a minimal training loop for local classifiers used in a hierarchical setup.

  1. Build or load an AnnData with appropriate obs columns describing labels.

  2. Use HierarchicalClassifier.run_feature_selection(node, …) to select features.

  3. Create a local classifier with create_local_classifier(node, classifier_type=…).

  4. Train with train_single_node(node, …).

Example:

clf.run_feature_selection('Root')
clf.create_local_classifier('Root', classifier_type='DenseTorch')
clf.train_single_node('Root', epochs=10, batch_size=32)

For hyperparameter tuning see Compocyte.core.tuner.