Is it possible to run the AI FTP detection over all my historical power data?
I like the AI FTP very much, because I hate all the all out tests.
The higher my FTP is the harder those all tests are.
I have 3 years power data, most outside. And started with TR in summer 2022.
I’m glad to see you decided to upload all that history into TrainerRoad.
Your most recent training has the biggest impact on assessing your current fitness and FTP, so consistent ride data within the last year would be the most relevant for getting an accurate AI FTP Detection result right now.
However, that long-range historical data does help improve TrainerRoad models across the board, for everything from workout recommendations to FTP predictions. Thanks for syncing all those rides!
Stoked you’re with us, let me know if you have any questions about AI FTP Detection or otherwise.
Many thx for your answers. My English is not that good, so I try to explain with the screenshot.
You see the black FTP change symbols. The last 6 are from the AI and works very well!
My question is, if AI could calculate all FTP changes from the years before. Let us say every month a calculation.
Greetings from Bavaria
So the ask is to get historical ftp levels so all the metrics that use ftp as part of the calculation can be more accurate? So less to influence what you are doing now but more to see your long term trends?
I like that idea. Even just the historical tss tr shows is kind of useless if your ftp is different significantly from the number it used to calculate tss. Then if you use more advanced metrics like what intervals.icu have can make it much easier to interpret the past
No problem! I think Im understanding that you’d like to know retroactively what your FTP was at previous points in time. While this is not a feature that we have in place currently, I’ll keep an eye on this thread as a feature request.
I was wondering if those old rides with inflated power were the reason aiFTP decided my FTP was going “up up & up” 250 -->265 → even 269 in Aug
vs how i felt (and ultimately tested) “not so fast” 235W.
Anyway, i’m sticking to 235W and not upping ftp until i can do Gray PL 5.9 (2x20min @235)
Nope! The reason that was happening was because you were nailing workouts and your survey responses were showing the workouts were Moderate in difficulty.
Once you started training less, you lost fitness, which is when training got more difficult and AI FTP Detection suggested 244. Considering the state of your training leading up to that test where you scored 235, that test result likely wasn’t a full representation of your current capabilities.
The main takeaway here is that when you were training consistently, you were getting faster and AI FTP Detection recognized that. When you were training less, you were getting slower and AI FTP Detection recognized that as well. In other words, consistency is key, and you can trust AI FTP Detection.
Let’s not mix Jul + first half of Aug (when i was riding 3-5 times/wk so with ok consistency)
with late Aug & Sept ( taper / IM race, travel & recovery).
The aiFTP problem occured in the first period.
Out of 4 structured workout in early Aug: I rated two Hard , 1 Very hard and then failed the 4th.
On Aug 11th i failed Cap Range PL 4.2 @ 264W and the next day Aug 12th aiFTP proposed a bump to 269W !!! That’s where aiFTP went up up & up. 244 would have made sense.
A few days later i did ramp test and got 235 which is maybe low as I typically undertest.
Afterward, i tapered, raced an IM and recovered.
aiFTP gave me 244W 3 weeks later (Sept 7th). I reverted it to 235 and built up the PL instead.
Very interesting this current communication! @IvyAudrain Does AI consider if using in the workouts 10 sec back paddling or lowering the intensity some percentage?
Or does it only consider the responses like very hard or all out?
(I work myself in the AI field of SAP support and see always asking customers: what is it doing? Well, in many cases not really to explain because is self learning…
The duration and number of those backpedals can contribute to an interpretation of ‘struggling’, but that can also happen when you need to tighten your shoes, reach for a bottle, pause to stretch, etc. This is why you’re given the option to clarify that you didn’t struggle if those backpedals weren’t related to the difficulty of the workout.
Adaptive Training asking if you struggled based upon backpedals/intensity decreases varies so much depending upon the frequency and severity of both, and there’s no one all-encompassing formula such as ‘x seconds of backpedal = struggle’, or ‘x intensity decrease = struggle’. It gets very complicated in the context of each individual workout and what exactly you’re doing, but the good news is: AT is wicked smart when considering that.
I would really like to be able to see what TR thinks my FTP would have been over time, even when I wasn’t using TR. One of the things I’ve always felt is a gap in my training is inconsistency in testing, so that it can be hard to tell what did and didn’t work.
I’ll add my vote for retroactive aiFTP as a feature. Would be neat to see what aiFTP thinks it was relative to where I tested at.
I think people forget or don’t understand with ai or machine learning that it’s all probability and model based. There’s going to be times where that prediction is off, but it seems like with adaptive training that even with a mis-estimated FTP that one’s training plan will get brought back in to line within a couple workouts. Even as a scientist that deals with probabilities and statistical uncertainty all day, when it comes to training I have a hard time trusting the system. If I could just stop medaling with aiFTP detection as soon as I’m eligible every time, I’d be laughing!