Originally Posted by
PeteHski
I was only referring to HR max in the sense that the whole HR range is widely different between individuals of the same fitness level. So I’m not sure how the model can adjust for such individual variation. Especially if it doesn’t even capture an individual’s complete HR range. For example if it measures my HR at 150 for a particular power output, how does it estimate where 150 is relative to my HR max or my LTHR?
I'm pretty sure from the description that Marco's algorithms don't really care about the complete range of HR; I'm guessing they look at something akin to watts/HR and how that varies across power and HR; the main underlying modeling assumption is probably that watts/HR isn't constant but changes smoothly. The algorithmic tricks are in finding the "right" lag or kernel over which to detect these changes, and modern ML methods are pretty good at brute-forcing this.