Principle:Kubeflow Pipelines Model Evaluation Metrics
Sources: Scikit-learn Metrics
Domains: Machine_Learning, Evaluation
Last Updated: 2026-02-13
Overview
The process of computing quantitative metrics that measure model prediction quality against ground truth values.
Description
Model evaluation calculates metrics like Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared by comparing predictions to ground truth. These metrics determine whether a model is good enough for deployment or needs further training. In iterative training loops, the MSE threshold drives the termination condition.
Usage
Use after model prediction to quantify model quality and drive automated training decisions.
Theoretical Basis
Regression metrics quantify the difference between predicted and actual values:
| Metric | Formula | Interpretation |
|---|---|---|
| MSE | (1/n) * sum((y_true - y_pred)^2) | Lower values indicate better fit; penalizes large errors |
| MAE | (1/n) * sum(abs(y_true - y_pred)) | Average absolute deviation; more robust to outliers than MSE |
| RMSE | sqrt(MSE) | Same units as the target variable; easier to interpret than MSE |
Threshold-based decisions automate the train-evaluate-check loop: if MSE exceeds the threshold, additional training iterations are triggered.