# metrics

Python API: `luxonis_train.attached_modules.metrics`

Metrics that compare the predictions of a node with the labels.

A metric attaches to a node through the `metrics` list of the node in the config. The trainer updates each metric on every
validation and test batch. At the end of the epoch, it computes and resets each metric and logs the results.

Mark one metric with `is_main_metric` in the config. When no metric sets it, the config marks the first metric. The trainer keeps
the checkpoints with the highest values of the main metric in the `best_val_metric` directory.

[MeanAveragePrecision](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/mean_average_precision/mean_average_precision.md)
and
[DetectionConfusionMatrix](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/confusion_matrix/detection_confusion_matrix.md)
consume boxes after the head's non-maximum suppression, so its `conf_thres` and `iou_thres` affect their results.
[PrecisionRecallCurve](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/precision_recall_curve.md)
uses the raw boxes and performs its own suppression.

To write a new metric, subclass
[BaseMetric](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/base_metric.md).
The example of
[MetricState](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/base_metric.md)
shows a complete subclass.

## Child Pages

 * [confusion_matrix](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/confusion_matrix.md)
 * [mean_average_precision](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/mean_average_precision.md)
 * [base_metric](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/base_metric.md)
 * [dice_coefficient](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/dice_coefficient.md)
 * [embedding_metrics](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/embedding_metrics.md)
 * [mean_iou](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/mean_iou.md)
 * [object_keypoint_similarity](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/object_keypoint_similarity.md)
 * [ocr_accuracy](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/ocr_accuracy.md)
 * [precision_recall_curve](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/precision_recall_curve.md)
 * [torchmetrics](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/torchmetrics.md)
 * [utils](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/attached_modules/metrics/utils.md)