evaluating-models

Evaluate machine learning models by analyzing MLflow history to compare best runs and track performance trends. The system automatically evaluates completed training runs, generating concise reports that summarize historical benchmarks and provide data-driven recommendations.

Confusion matrix analysis identifies class-level errors and misclassifications in your machine learning model. By evaluating completed training runs, the system generates reports that highlight specific class performance issues and provides actionable next steps to address them.

Interpreting PR curves and F1-scores involves analyzing precision-recall trade-offs across different classification thresholds. The evaluation system processes these artifacts from completed training runs to generate objective performance reports with actionable next steps.

Comparing multiple training runs to find the best model performance is supported through historical benchmarking via MLflow. The system applies cross-run comparisons, referencing the best_run to generate data-driven reports that summarize performance trends.

If model evaluation artifacts are missing, the evaluation system guards against incomplete data by ensuring comparisons and reports account for absent artifacts. This prevents failed evaluations and provides actionable next steps based on available training run data.

Completed training runs are required before starting model evaluation, as the system analyzes resulting artifacts like confusion matrices and PR curves. Apply the evaluation workflow to these completed runs to generate concise, data-driven performance reports.