@Nate_Pearson - as someone who is really familiar with Monte Carlo Simulations and there usage, will the sensitivity analysis from the Monte Carlo Simulations be shared with the end-user (or at least have a super advanced option for those who understand to be able to see this?)? This is really the crown jewel benefit: what training structure lever (e.g., volume, work to rest interval, plan structure components, etc.) does the new AI think has the most bang for the buck to increase FTP? And does 1 lever (e.g., volume) massively outweigh the contribution of all others, and where is the break point (e.g., going from 8 - 10 hours gets me 80% of the benefit from increasing volume, while going from 8 - 12 hours only gets 87% of the increasing volume benefit, etc.)?
I use them all the time as well, and was wondering the same thing. The thing im curious about is the data structure. Are all the potential correlations between the predictor parameters accounted for in the sampling or are they assumed to be independent? Presumably TR can do the former since they have all the data
The other thing that would be potentially really helpful for future product development would be a value-of-information analysis, where instead of cost-benefit you could do “harm-benefit” (say “Overtraining episodes per % FTP gain”) and identify which parameters are contributing the most to uncertainty, and then put resources into better metrics/ways of reducing that uncertainty.
Does that mean swims will be imported and shown on the calendar soon too?
Wish i had this during ironman training last year - would have been a godsend. Five months post race of no exercise after significant burnout and only just going back to it now - 3+ stone overweight…
Im not sure yet.
If you’d like an alpha tester, ping me
If you / TR decide not to do this, I get it as explaining this to folks who don’t understand / haven’t worked with Monte Carlo Simulations will be super difficult
We have a project to show swims on the calendar along with other activities. It won’t be as feature rich as cycling, but they will be there.
It’s in development now and I don’t have a launch date yet.
Based on the papers I get asked to review, a lot of people who ostensibly have worked with it don’t understand it.
I can believe that
For teaching purposes, I’ve found using election predictions seem the most intuitive. I’ve got some screenshots from the 2016 primary that show the distributions of poll results, and say “Candidate X is favored 96% of the time”–most people are used to the idea of “margin of error” with polls, and it’s easy to see where the tails of the distributions overlap.