Module metrics
Source - ClassMetrics
- Per-class metric report entry.
- ConfusionMatrix
- NxN confusion matrix. Entry [i][j] = count of samples with true class i
predicted as class j.
- Average
- How to average per-class metrics for multi-class problems.
- accuracy
- Classification accuracy: fraction of correct predictions.
- argmax_classes
- Compute argmax along the last axis, returning class indices.
- classification_report
- Per-class precision, recall, F1, and support — like sklearn’s classification_report.
- f1_score
- F1 Score — harmonic mean of precision and recall.
- mae
- Mean Absolute Error: mean(|y_true - y_pred|).
- mape
- Mean Absolute Percentage Error: mean(|y_true - y_pred| / |y_true|) * 100.
- perplexity
- Perplexity from cross-entropy loss: exp(loss).
- perplexity_from_log_probs
- Perplexity from a flat array of per-token log-probabilities.
- precision
- Precision for multi-class classification.
- r2_score
- R² (coefficient of determination).
- recall
- Recall for multi-class classification.
- rmse
- Root Mean Squared Error: sqrt(mean((y_true - y_pred)²)).
- tensor_accuracy
- Compute accuracy directly from logit tensors and one-hot/class-index targets.
- top_k_accuracy
- Top-K accuracy: fraction of samples where the true class is in the top-K predictions.