tests/README.md

Tests Directory

Automated test and validation scripts for the AI Battery Lifecycle Predictor.

Test Scripts

1. test_v2_models.py — Comprehensive V2 Validation

Validates all 14 v2 classical models against test set with detailed reporting.

Usage:

python tests/test_v2_models.py

Output:

• artifacts/v2/results/v2_model_validation.csv — Full metrics table
• artifacts/v2/results/v2_validation_summary.json — Summary statistics
• artifacts/v2/figures/validation_accuracy_bars.png — Accuracy ranking chart
• artifacts/v2/figures/r2_vs_accuracy.png — R² vs accuracy scatter plot
• artifacts/v2/figures/best_model_analysis.png — Best model performance
• artifacts/v2/figures/per_battery_accuracy.png — Per-battery accuracy heatmap
• artifacts/v2/results/v2_validation_report.html — HTML report
• artifacts/v2/results/v2_validation_report.md — Markdown report

Target Metrics:

• Within-±5% SOH Accuracy ≥ 95% (primary success metric)
• Within-±2% SOH Accuracy (secondary)
• R² ≥ 0.95 (correlation)
• MAE ≤ 3% (mean absolute error)

2. test_predictions.py — Quick Endpoint Validation

Quick validation of prediction endpoint with sample features on model registry.

Usage:

python tests/test_predictions.py

Output: Console output showing predictions for 4 test scenarios:

• Early life cycle (SOH ≈ 99%)
• Healthy cycle (SOH ≈ 97%)
• Degraded cycle (SOH ≈ 80%)
• End-of-life cycle (SOH ≈ 40%)

Running Tests

Run all tests

python tests/test_v2_models.py
python tests/test_predictions.py

Run individual test

python tests/test_v2_models.py    # V2 model validation
python tests/test_predictions.py  # Endpoint test

Test Data

Tests use:

• Features: artifacts/v2/results/battery_features.csv (2,678 samples)
• Target: SOH (State of Health) percentage
• Split: Intra-battery chronological (first 80% cycles → train, last 20% → test)
• Batteries: All 30 usable batteries in both train and test

Model Versions Tested

Model	Category	V2 Status
ExtraTrees	Tree	✓ v2.0.0
GradientBoosting	Tree	✓ v2.0.0
RandomForest	Tree	v1.0.0
XGBoost	Tree	v1.0.0
LightGBM	Tree	v1.0.0
SVR	Linear	v1.0.0
Ridge	Linear	v1.0.0
Lasso	Linear	v1.0.0
ElasticNet	Linear	v1.0.0
KNN (k=5, 10, 20)	Linear	v1.0.0

Success Criteria

✓ Model passes if: Within-±5% Accuracy ≥ 95%

Current pass rate: See latest report in artifacts/v2/results/v2_validation_report.html