What problem does it solve? Setting up a correct data preprocessing pipeline for battery State-of-Health (SOH) prediction is error-prone: wrong train/test splits, scaler leakage, and inconsistent seeds silently corrupt model results. This Skill scaffolds a standardized preprocessing module that enforces the NASA battery split convention and prevents data leakage. ## Core Features & Use Cases - Standardized NASA split: Hardcodes train batteries B0005–B0007 and splits B0018 into 70% validation / 30% test by timestep order. - Leakage-safe scaling: Fits MinMaxScaler only on training data, persists it to models/weights/scaler.pkl with version metadata, and provides a load_scaler() function for inference. - Reproducible windowing: Builds sliding windows of 30 timesteps over voltage, current, and temperature features with a fixed seed of 42. - Use Case: A machine learning engineer starting an SOH prediction project runs the scaffold to get a ready-made data/preprocess.py that loads battery CSVs, produces (N, 30, 3) feature arrays, and saves a versioned scaler artifact. ## Quick Start Ask the AI to scaffold the preprocessing pipeline for the NASA dataset, or specify CALCE to get prompted for battery IDs before file generation.
