Event detection accuracy plot #3
Description
The context is time-series ML where we build models to detect (point or interval) events.
When only graphing the time-series, overlaid with the event (detected or actual), I often see something like this:
for point events, and for interval events like follows:
But it can also be useful to overlay both the actual event location and the detection location.
This allows one to get a visual intuition not only of the accuracy, but how/when the detector fails (e.g., it's off by a constant shift, or it's off when the sound is loud, etc.).
What I've seen to do this is the same graphs as above but with two event location kinds of different colors.
I think (but am not sure before I see it) that instead of doing that, having them both stretch from min to max of the y-axis, have one stretch from the min to the mid and the other from the max to the mid.