CheckpointMetricType
Enums · AiDotNet.Enums
Standard metrics for checkpoint selection and early stopping.
For Beginners
Use Loss for most cases, or Accuracy for classification. The metric determines when to save checkpoints and when to stop early.
How It Works
These metrics determine which checkpoint is considered “best” during training.
Fields
| Field | Summary |
|---|---|
CheckpointMetricType Accuracy |
Accuracy metric (higher is better). |
CheckpointMetricType BLEU |
BLEU score (higher is better). |
CheckpointMetricType Custom |
Custom metric defined by user. |
CheckpointMetricType F1Score |
F1 score (higher is better). |
CheckpointMetricType Loss |
Training or validation loss (lower is better). |
CheckpointMetricType Perplexity |
Perplexity (lower is better). |
CheckpointMetricType ROUGE |
ROUGE score (higher is better). |
CheckpointMetricType RewardScore |
Reward model score (higher is better). |
CheckpointMetricType WinRate |
Win rate against reference (higher is better). |
int value__ |