Pytorch Training Workflows

Repository ML Priorities

• Keep rotational transmission error as the target quantity.
• Preserve the distinction between analytical structure and ML-based compensation.
• Keep operating variables explicit: input speed, applied torque, oil temperature, encoder zeroing, and DataValid windows when relevant.
• Keep artifact layout, run_name, and run_instance_id semantics correct.
• Favor inspectable intermediate stages over opaque shortcuts.

Implementation Pattern

Prefer this sequence:

1. Read references and local implementation.
2. Identify behavioral constraints and affected workflows.
3. Check library API details with Context7 when version-sensitive.
4. Implement the narrowest safe change.
5. Verify training, smoke-test, validation, and documentation implications.

File Targets To Read First

• reference/
• doc/reference_summaries/
• doc/running/active_training_campaign.yaml
• scripts/models/
• scripts/training/
• config/training/
• output/registries/