I am sure your predictions will be good for the vast majority. Was more focusing on how with ML when the model doesn’t predict the real results there can be something worth looking into. (when the numbers are signifigantly off, not a rounding error off) Not because someone complains, but more the automated backend just looking at what they actually did vs the prediction of what they should have done. Then when you have a data set of those that predicted too high and those that predicted too low you can look into general trends of what is special about them. Some of those data points should be thrown out because the change is based on things you don’t know about (bike fit change, different equipment, etc) and things you might be able to predict but still easy to ask (they feel tired or worn out) so could be very useful to have a survey for ramp tests.
Assuming always good manual RPE responses from surveys seems like a bad assumption as it is almost the same as assuming someone complaining their predicted FTP is right. I’m not saying to go off of them contacting support or complaining in the forums, just based on the data you see.
My prediction is that those who are newer to training and/or starting at a lower power will be harder to predict as their rate of increase will be greater then those who have been training awhile or already high in power. They are further from their bodiy’s potential and have more areas where they can improve (the different types of physiological improvements like mitocondrial growth, capillary beds, etc) but will be hard to know how fast they can improve. Go under in your prediction and they won’t improve as much as they can but if you go over in your prediction you might get closer to overtraining. I’m guessing being new to it also means they won’t be as good at rating their RPE for different workouts
Wait to see if anything major comes up from the podcast today. Only one big thread at a time or else the split attention will keep the thread from exploding. If nothing today, wait till 7 days from the last exploding thread. Need some recovery first. Did you thinking this was only for training to be stronger on the bike?
I’m not saying the algorithms are going to be the same, but before I switched Wahoo, my Garmins always gave me FTP estimates. I never had any reason to argue with my estimate FTP. It was nice not testing and I could take any workout from TR, TP, or even custom FTP workouts from private coach’s trial plans and finish them.
That said, there were times when I’d do a series of rest/recovery rides and my Garmin would knock my FTP down. I didn’t have the option to tell it “I wasn’t trying hard, you donkey.”
My policy on TR is likely to be the same it was on Garmin. If I am sidelined for a significant amount of time and I believe I’ve really lost fitness, I’ll accept a suggested decrease. Otherwise, ignore decreases and accept the increases. That sounds stubborn and stupid but I know from past experience, if you ignore a drop and then 2 weeks later are setting a new PR FTP from the same estimator, you can feel confident you were right to ignore the drop.
PS: Thank you so much for this new tool. I’m so excited - I thought it would be a lot further out. I’m still going to do another ramp test tho bc I’m excited to do apples to apples at the end of my first block of base training.
This is also what intervals.icu does, based on Mortons 3P model.
I’ve tested once in 3 years, and all that test told me was what I already knew. Estimated FTP is the way forward, and the positives way outweigh any negatives in my opinion.
For the vast majority I can’t see how the positives wouldn’t outweigh the negatives.
For me, and I’m in no way exceptional so I would assume this applies for many, I don’t need my FTP setting to be perfect, especially with adaptive training. As long as it is in the right ballpark, that will be a good starting point. If I find that workouts are a little easy, maybe because the FTP setting is a little low, adaptive training will serve up ever more difficult workouts. If the setting ends up a bit high, I’ll struggle with workouts and adaptive training will adjust things. If the FTP estimation system then builds on this to adjust that base level off the back of the actual performances, then that surely is going to give great results for the vast majority
My guess is that the vast majority of TR users “under test” on the ramp test. Total without data guess, but my thinking is that most people cannot really push themselves to go “all out” - yes, there will be exceptions. So I think TR’s use of 75% of the ramp test is probably safe for the vast majority of the TR user base to get a number that is “good enough” to use for “threshold” and below intervals.
And the flip side of that - those that can go all-out may over test when using 75%. Coach Chad’s deep dive on VO2max touches on the fact that different people have different ftp as % of max aerobic power, and that max aerobic power is what you get from the final moments of a ramp test.
These are all reasons why it is a great idea that TR is attempting to do a separate FTP estimate, hopefully using both a ramp test and subsequent (longer) intervals in workouts to triangulate a same-or-better FTP estimate.
I would say that no matter how good the estimation will be, there should be one improvement - overtesting. I am always amazed how many FTP tests TR want you to do:) I know it’s needed for new users are the improvements are pretty rapid but for more experienced people I cannot imagine they need a test in almost every block. But this is only my personal view and maybe I am biased.
I’m not disagreeing with either of your points: yes, there are people who “over test” (their FTP as % ramp test is less than 75%, so the ramp test over estimates their FTP); and yes, the new FTP estimator is a good thing. I really want to test out the estimator myself
My point was that for the vast majority of the TR user base, the ramp test (for a test) is an okay method to create a number to base “threshold” & sub-threshold workouts on.
The question should be: I’m curious to see how TR’s AI FTP Detection compared to WKO5 and Intervals.ICU for getting me into the correct next workouts for me.
Pick any definition of FTP and derive it from any method. Now, what should be your specific Vo2 max repeats be at, for how long, and for how many intervals?
In our opinion, you need the combination of a good FTP and the proceeding workouts at the correct levels for you.
It most likely won’t work this way in most cases :-D. We’re using real data and history to generate this. So unless in real cases athletes who go from your volume down to another volume and intensity combo actually did lose fitness it shouldn’t lower you.
Another way to put it you can get an AI FTP increase by only doing sweet spot work. You don’t need to go all out.
Tr is going down the path of using ai to predict what you can do based on what you did in the past. So it tries to figure out if you did certain workouts your fitness will improve in a certain way so it can adjust pl levels and soon ftp.
Other companies are mostly about different ways in measuring your max power and calculating your numbers based on that. Trainer road may also do that in order to better train their model but can still predict without it.
Agreed. We will find out this week, as I’m switching from doing long (I was doing 2x40 @90% FTP with 5 minutes rest) to doing shorter - exact is TBD after talking with my coach on Monday - intervals but at 95% - 97% of what will be a new and higher FTP.
