evaluate-model

Model evaluation measures performance on test datasets by computing metrics like accuracy, precision, recall, F1, and AUC for classification tasks, or MSE and MAE for regression. This identifies whether your model generalizes well or is overfitting, and provides a confusion matrix and per-class breakdown for detailed analysis.

Metric selection depends on your task type. For classification, use accuracy for overall performance, precision and recall for class-specific trade-offs, F1 for balanced scores, and AUC for threshold-independent comparison. For regression, use MSE or MAE. Baseline comparisons help contextualize whether one architecture outperforms another.

Yes. Overfitting occurs when validation metrics are significantly higher than test metrics. Evaluation reports showing both sets side by side reveal this gap. Large divergence between validation accuracy and test accuracy indicates the model memorized training patterns rather than learning generalizable features.

Evaluation reports include task type classification, per-class performance breakdown, confusion matrices showing prediction errors, baseline comparisons to establish performance context, and error analysis highlighting where the model struggles most. These components support debugging and architectural decisions.

Yes. Model evaluation requires labeled test or validation datasets with ground-truth targets. The Skill compares your model's predictions against these labels to compute accuracy, precision, recall, and other metrics that measure how well predictions match reality.