I just listened to the two AI podcasts and I am super stoked. I am still a fairly new cyclist but I’m a researcher, programmer and data geek, so I’m very curious and interested in the whole thing. I have some questions, thoughts and suggestions.
Questions
Why did FTP change?
One thing about the podcasts that confused me was that idea that FTP had to be corrected to a “level 3” athlete level. I guess I don’t understand this. As discussed on the podcast, FTP is a functional benchmark. Why do athlete levels come in to play at all? Was the FTP reported by trainerroad not really the true prediction for FTP?
What exactly is the model predicting? What is the “right” workout?
Based on my understanding of the podcast, the model is predicting the outcome of a single future workout and the perceived difficulty. But there is also a lot of talk about the “right” workout. How is that decided? What is being optimized for?
Thoughts / Ideas
FTP as a metric
Many people are used to tracking their fitness by FTP, so I understand reporting that as a benchmark. But it would be cool if athletes could track their own custom benchmarks. For example, let’s say I really care about my ability to sprint. It would be cool if I could select a workout that tests 20 minutes of intervals and see what power level the model expects me to be able to accomplish for those intervals. FTP is sort of like this for a uniform one-hour ride.
PCA for summarizing fitness
As an alternative to using FTP to summarize fitness, it would be interesting to use something like principal component analysis (PCA) to try to summarize fitness. We basically use FTP for this. But maybe the model knows a better metric than FTP. Or, maybe it would be better to have a multi-dimensional metric that represents fitness (probably corresponding to abilities at various power levels).
Exponential search
Nate said something in one of the podcasts about finding an ideal workout schedule being exponential. How is the workout schedule selected now? Could you use something like beam search (if you aren’t already)?