ml-engine-pipeline

You generate an sklearn training pipeline by building a classifier with a predict_proba interface and a matching _config.json file. This Skill ensures engine compatibility by validating probability shape, sum constraints, and LabelEncoder class ordering before deployment.

You ensure consistent feature ordering by defining a FEATURE_SCHEMA as the single source of truth. This schema enforces identical feature sequences across the sklearn training pipeline and the inference engine during deployment.

Yes, you can train an ML classifier for IMU and PPG windowed sensor datasets in Google Colab. This Skill generates ready-to-run notebooks that build pipelines supporting stat, derived, cross-signal, and meta features for activity classification.

Your ML model fails deployment due to a mismatch between the sklearn LabelEncoder class ordering and the inference engine's expected probability shape. This Skill performs pre-save compatibility checks to verify probability sums and prevent these deployment contract errors.

The best way to validate an sklearn model before production inference is to run pre-save engine compatibility checks. This verifies the predict_proba interface, output shape, probability sum constraints, and feature count against the inference engine's requirements.