Description
The tune metric specifies the key metric that AutoML uses to optimize training experiments. This metric plays a critical role in:
- Early Stopping: During training, Kumo monitors the tune metric to decide when to stop training models that are unlikely to improve further. Learn more about this feature in the
early_stoppingdocumentation. - Model Selection: Among all models trained during AutoML, Kumo selects the best-performing model based on the tune metric. This selected model is then evaluated on the test set for final validation.
The tune metric needs to be in the list of metrics in the model plan. For available metrics, please refer to metrics and Kumo Evaluation Metrics.
Supported Task Types
- All
Default Values
| Task Type | Default Value |
|---|---|
| Binary Classification | auroc |
| Multiclass Classification | acc |
| Multilabel Classification | ap_macro |
| Regression | mae |
| Forecasting | mae |
| Temporal Link Prediction | map@10 |
| Static Link Prediction | map@10 |
| Multilabel Ranking | map@10 |
Updated 23 days ago