Average Precision

https://www.kaggle.com/code/metric/event-detection-ap/notebook

• Use this metric when you want to see how well your model identifies (not predits!) “events” in a time series
• Here’s how it works:
  1. Predicted events are matched with ground-truth events.
  2. Each group of predictions is scored against its corresponding group of ground-truth events via Average Precision
  3. The multiple AP scores are averaged to produce a single overall score.

Note: This metric could be abused:

Note: if you submitted multiple predictions for the same sleep event, you’ll get the same score (or better)

Link to original

• you can abuse this to get a better score