I’ll defend TR but I usually try to point people in the direction of looking into through blog posts or whatever or then getting ahold of support. Humans have a way of Complaining and when they get their answer they’ll find something new to complain about. Now I’m complaining about complaining, I’m not any better. In the end I try to share TrainerRoads information that they have online. Usually it answers questions and if not support hopefully will.
1 Like
I think this is an accurate assessment of it’s limitations. Endurance, tempo, sweet spot, and Threshold are all zones that are based of % of FTP and not all out efforts that are FTP dependent - so in my opinion when your FTP increases those zones should not decrease - maybe decay over time like they do now, but they are not tied to FTP as close as Z1-Z4.
1 Like
I largely agree with you, but as someone who really struggles with threshold (vs V02 for example), I definitely need the threshold PL to change when the FTP steps. If TR used a format where your threshold floats daily rather than steps every few weeks, lower zone PLs wouldn’t have much use and wouldn’t budge much at all. But I think I prefer the way they have it set up with fixed/stepping FTP and floating PLs.
1 Like
Agreed - if you have a bump, bringing threshold down will help you complete those workouts (a 2.0 vs. a 5.0 for instance), but you can do VO2 at any time much higher than a 1.0 - if you’ve done them before and have experience with structure because they’re not tied to %FTP and should be done in resistance mode.
Thank you for coming to my ted talk.
*Cunningham’s Law
Lol, had to do it, it’s the law.
2 Likes
1 Like
TrainerRoad !!!
I think you are misreading what I wrote. I meant that they are made to work hand-in-hand even though they are independent modules. AI FTP was mentioned when AT was first unveiled as a part of TR’s ML efforts.
Don’t take my word from it, just re-listen the relevant part of TR’s podcast where AT was first unveiled: @Nate_Pearson had a section explaining that their intention is for athletes to never have to take an FTP test ever again — he was describing AI FTP without calling it as such.
Yes, you can use both independently (I like ramp tests, so I am going to keep doing them for now), but it is part of a larger vision for TR’s future and they have been designed by TR to work in tandem — even though you can use them separately.
The fact that you can use both separately is arguably an advantage: even if you e. g. don’t currently use AT, you can use AI FTP. Conversely, I use AT without AI FTP — I like doing ramp tests. 
Minor nitpick: AI FTP does take outdoor rides into account, AT currently does not.
I think you need to separate how well a ML algorithm works from how simple it is. ML isn’t magic and needs qualified people to implement. E. g. how do you evaluate how well an algorithm works? What variables do you include and what variables do you exclude? This is all quite tricky, because it requires domain expertise (here: exercise physiology) and expertise in ML.
Overall, it seems to me that you perceive something as simple when “all it does is adjust some progression here and there.” AI FTP is “even simpler” from that perspective: it just spits out a number in the end. Does that mean it is simple? I don’t think so.
TR has released stats that their athletes e. g. have lower failure rates when they use AT, are more consistent and have better outcomes. That doesn’t mean it works equally well for everyone or that it doesn’t have any flaws. (I can list some if you want apart from outdoor workouts not being included at present.) @Nate_Pearson also made public that they have an overall fitness score (in addition to plenty of other markers). The difficult part IMHO is how to judge success. No doubt that future evolutions of AT will revisit that (by e. g. emphasizing certain metrics over others depending on the plan that you chose).
No, a linear regression is not Machine Learning.
No offense, but I think you fundamentally don’t understand what ML is and how it works. You seem to think that something is simple if the output is simple. AI FTP is a complicated function that computes a single number. The difficult problem is finding that function from statistical data without knowing it in advance.
So you could also have a complicated algorithm (whether it is based on ML or not doesn’t matter for the moment) that determines ramp rates. Still, just because the output seems simple to you doesn’t mean the function is.
Why does AI FTP seem more advanced to you?
I think that’s quite simplistic, and is not really what this discussion is about. IMHO we need to tease out whether the premise is “Why does AI FTP work better than AT?” or what @patrickhill means by smarter?
You can go into the technical details (to the degree that they have been disclosed publicly, of course) without being a fanboi. E. g. opining that you shouldn’t use “smarter” in this context is different from claiming that either one or both work wonderfully for users. (When I used AI FTP last after a 4-week hiatus, its estimate was way off.) Ditto for pointing to TR’s public resources.
I took the liberty of emphasizing two words in your post: I completely agree with how you phrased statement, because I just wouldn’t say that one algorithm is “smarter” than another. The latter suggest the techniques used by AI FTP are more sophisticated, but that is not the case. You can have a complicated solution to a problem that doesn’t work very well …
If we want to discuss how well AI FTP works compared to AT, that’s a great discussion to have.
Except that in this case it arguably does: AT has been designed from the ground up with AI FTP and scoring of outdoor workouts being part of the overarching vision. These were known limitations of the first release of AT, and TR works on filling in the missing pieces. AI FTP is out as a public beta and outdoor workouts is in internal beta. The fact that both provide utility on their own is an added bonus.
PS I hope that in the future AT will not just look at PLs of individual workouts but of time in zone across a week or so: I often tack on 30-minute endurance workouts, but these have ridiculously low PLs. They don’t raise my endurance PL even though they arguably raise my fitness. But I digress.
Linear regression is a fundamental algorithm in the field of supervised machine learning.
Y’all have lost me on whatever it is you are debating.
8 Likes
I don’t want to sidetrack the discussion too much. Linear regression is an elementary technique from statistics, which, yes, is used in machine learning. There are also plenty of sophisticated purely statistical approaches to Big Data (a keyword, which has been replaced by ML in public perception at least), but teasing those differences apart gets tricky.
However, when someone who with very high probability is a non-expert in the field (an assumption I make) uses words like “linear regression”, I think that they think of the simple tool from statistics where you plot 2d data points and “draw a best-fit line” through them so as to minimize the deviation. I could be wrong here, but I think it is reasonable to make that assumption.
1 Like
Elementary? It’s a foundational algorithm and can be multivariate / multidimensional. I’m done with the debate. Good luck 
3 Likes
I work in ML, I think it’s very easy to have scenarios where the AI FTP model is more appropriate than AT. Even if they both use the same raw dataset to train. The two programs have fundamentally different outputs: one is an single FTP number, the other is a set of adaptations. Thus making them very different things.
Anyways, knowing how it functions doesn’t help at all in this case. If one tool is more accurate and helpful then I think it make sense to base your training on that in this instance. That said, I’m doubt these micro tweaks make a big difference in the big picture, so unless AT is waaay off, I’d just do what it says and save on mental load.
7 Likes
To me it is elementary, the first fit algorithm I have learnt when analyzing data. I have seen plenty of abuse (including applications to plots with two data points 
in a poster in the geology department, where I had an office for a while, or the obligatory point clouds where no linear dependence between the variables is evident), too.
Why? Unlike me, you have proper expertise in the topic, and I think your opinion would be valuable.
I reckon that scoring is much simpler with AI FTP than with AT, so from that perspective at least, AI FTP seems to be the far simpler problem. On the other hand, it could also mean that this explains why some people in this thread think AI FTP works better than AT.
Which I think brings it full circle to where this all started… AI FTP feels more like it comes from deep data analysis and it is eerily accurate (from a TR workout perspective) in many cases; whereas PLs feel like they come from some simple standard algorithms and many people (I’m guilty) end up gaming the AT system to get to accurate PLs more rapidly after a new FTP is assigned.
6 Likes
AT definitely doesn’t get PLs to what I think is the right level. VO2 are always too easy if I don’t do them for a while due to the natural decay they built in.
1 Like
Honestly, I think it is the opposite way around: I reckon AI FTP is a much simpler problem because it is very clear what you are optimizing for and how to score that.
With AT that is much less clear, because you can define success in many, many different ways and weigh them differently as well. E. g. does AT take specific aims of the various plans into account and e. g. give more emphasis to short power in the crit plan? I don’t know. This is probably one of the most closely guarded secrets by TR.
I’m not quite sure what you mean by gaming. Do you just mean that when e. g. energy systems you haven’t trained in a while (and PLs have decayed to ridiculously low values) that you will do “breakthrough workouts” that simply reset the PL to a more sensible value?
If that’s what you mean, then I am doing that, too. (Two common ones being endurance PL and currently, my sweet spot PL: I have been replacing my sweet spot Sunday workouts with endurance rides.) Although, I wouldn’t call it “gaming the system”, you just understand how it works and work within the current limitations.
Again, that doesn’t mean that AT is “simpler” or “more primitive”, it just means that AT as it currently works has quite a few limitations that will need to be addressed in the future. IMHO AT is simply trying to solve a much more complicated problem than AI FTP.
PS My experience with AI FTP is quite mixed. During regular blocks the predictions are fine (I wish I could compare with ramp test results afterwards. But it had significant trouble estimating my FTP after a longer training pause, it was off by 15–20 W. (My lactate threshold had decayed less than expected, instead, my endurance went to shirt.)
I’m with you @Pbase. I often use use AT’s suggestions, but i definitely game it. Sometimes it’s off because it missed some outdoor rides, or it decayed my levels too fast, or has my anaerobic PL down at 1.3 when my V02 PL (which is sometimes set by doing intervals in the anaerobic zone) is at 7.
AT and AI FTP detection are great, but the real genius is in the PL algorithms. Knowing if one workout is harder or easier than another and by how much is an amazing luxury. It makes AT possible and allows us to fine tune our training.
6 Likes
No, I don’t think so. No need to believe in marketing, just listen to what they say and compare it with reality. When they made AT public, the TR team told us to anticipate AI FTP, and now that feature is public beta. They explained how they see it as an integral part of a ML-based training platform, and at least the logic makes sense to me.
To be honest, I think you mistake my enthusiasm for interesting stuff happening at the intersection of different fields of science with enthusiasm for TR. 
I am curious to see what interesting stuff will happen.
Maybe there is no need in your mind, but in this case history tells us it was.
I think that fundamentally misunderstands what PLs try to quantify, your endurance in a zone/the number of matches/whatever you want to call it. This is distinct from your lactate threshold. Your FTP = lactate threshold may be set correctly, but your endurance may be higher or lower.
Yes, you can use PLs to accommodate changes in FTP (which is not a fixed number for 4–6 weeks, but changes anyway). But that’s not what they are fundamentally there to do.
When it comes to FTP, maybe. And this shows one important point: users don’t and shouldn’t care how a device, an app or a platform estimates things like VO2max or your FTP. I don’t know whether TR is any better or worse, because they use ML and a very big, unique dataset. Personally, I don’t care about FTP estimation, I like ramp tests and if they are an option, I’ll probably keep doing them. I don’t know whether any of the other platforms I currently use can and estimate my FTP in some fashion. Probably the biggest challenge in even deciding whether these different platforms have better or worse algorithms is that we need to sample a large cross section of people.
But when it comes to adapting training plans, I am only aware of one other competitor that uses methods from ML (thanks to these forums), but I have no experience with them. So I cannot say whether they work any better or worse. If you go broader and do not restrict yourself to platforms that leverage ML, you can include platforms like Wahoo’s SYSTM (spelling?). Again, I don’t know if they are better.
I think you fundamentally my enthusiasm for science and ML-based methods for TR fanboi-ism. I’m very enthusiastic about ML, because I have seen it grow up in my vicinity. E. g. in the late 2000s my best friend’s Master’s thesis on computer science was using Big Data techniques to improve automated translations for software, partially developed during an internship with a certain Fruit Company. About 8 years ago I had the privilege to meet the who-is-who in the mathematical theory behind it at a research institute, go to lunch with them several times a week and absorb knowledge by osmosis (that doesn’t make me an expert!). One of the participants was offered a six-figure consulting gig with a MLB team to use ML to analyze player movements. If she had wanted to get out of academia (she’s a professor now), that would have been a good ticket.
A few years later it was instrumental in the search, discovery and imaging the event horizon black holes, a discovery I am sure will be awarded a Nobel Prize in the future. (There is an excellent talk on Youtube by the leader researchers of one of the competing groups for the ML algorithms that were used to reconstruct the image data, Katie Bouman.](https://youtu.be/UGL_OL3OrCE).) If I sound enthusiastic, that’s because I am, it lies at the intersection where I am at home, the intersection between mathematics and natural sciences (I’m a mathematical physicist). However, ML is used even in small ways. I’ve seen more and more physics papers from experimentalists who have used ML-based methods in their data analysis. (Often senior researchers would answer me “my Master student did this and I don’t fully understand it …” when I asked about it.)
So you asked whether I think “TR can do eventually do a better job with their vast database?” Yes, absolutely, if they play their cards right, they can leapfrog the competition. My reasoning is simple: they are the biggest platform of its kind (Strava does not have all the information and cannot recreate a lot either), which gives it a strong platform effect (in the same way that Google dominates search or FaceBook has eliminated a lot of other social networks). The second big factor is the quality and amount of data, which is also unique. The smaller competitor (I have forgotten the name) has to optimize its algorithms with far less.
The biggest limiter I am sure is talent. Some friends and former colleagues who went into industry are working on applying ML to all sorts of problems, some as slimy, yet lucrative as “predicting a customer’s journey” (aka ads, this was a quote from a friend who got offered a job with Japan’s biggest ad company; he declined). Others are much more interesting and can rely on cutting-edge research (e. g. image recognition for satellite imagery). Modern APIs are at a point where many people can do the “mechanics” with shake-to-bake solutions.
But problems like AT require extensive domain knowledge and knowledge of methods in ML. This I am sure is the limiter in TR’s case. Clearly, they have attracted plenty of people with domain knowledge in cycling, finding the few people at the intersection is hard. I am not sure how many people like @WindWarrior are out there on this planet, but I am going to bet it is very, very few. Perhaps you can count their number on two hands, and I am not exaggerating (speaking from experience in research, the more you want depth of knowledge, the fewer people there are).
In a sense, ML is relatively simple: it is the search for a local minima or maxima of functions in a vast, vast space. The space is so vast that in many cases you wouldn’t have enough memory even if you made the entire universe into a computer (that’s a way of saying that the number of configurations is larger than the number of atoms in the universe). Technically, the biggest challenge is finding the right optimization function, i. e. a way to quantify how much better or worse algorithm A is compared to algorithm B. FaceBook and Youtube famously want to maximize engagement with their selection algorithms for items in your timeline or in the list of suggested videos. You can see the results very clearly.
1 Like