AI Training - Will it work?

But the TSS and IF are not the same in this or any other similar example… this did make me laugh during the podcast, smoke mirrors and marketing. It was a statement that hadn’t been thought through imo, because it clearly is not true.

PS That isnt what was said.

Sorry but this is false. TSS for 5 x 10 and 1 x 50 are the exact same for these intervals. Factoring in the rest periods in a 5 x 10 workout will likely give more TSS, but if you really want to compare apples to apples, you would have to add rest time and intensity to the end or beginning of the 1 x 50 workout…giving the exact same TSS.

IF will behave the same way.

Fair point. If something like that happens to me eg. 6x4min VO2max is too hard, I’d either just skip one or two intervals at the end or choose one 6x3min. I agree that there is a benefit if I don’t have to search for it in the library.

Ehm… you can’t just throw data at ML and it will tell you all what you want to know.

If you want to make a model based on your power numbers you’d need regular all-out tests in various time ranges. That’s what WKO4/5 does for years. It suggests you interval duration and power based on your PDC. However, if you don’t maintain your PDC it obviously won’t work. This might be different for others but I find it hard to integrate 3-5 different all-out sessions every month just so the algorithm can tell me what I know by doing the workouts anyway.

Serious question. How do they know about your performance?
Do you mean FTP derived from a ramp test?
Analysing a power duration curve?
Power file from a crit and the result?

I was refering to features of the available dataset not the amount of workouts.

Agreed, unless you have built a model at which you can throw data, then yes, you can throw data at it and it will tell you exactly what you ask of it

Again, one of the points of the AT they bring up in the podcast - there is no need for these periodic assessments, after which your workouts are static/benchmarked until the next assessment. How long is an all-out assessment accurate? A day? A week? Not much more for sure. Thus, using more advanced data analysis to track your progress and fitness continuously is useful. As they repeated several times on the podcast, the right workout on the right day based on your entire training history. Not just your most recent PDC, ramp test, 20 min test, or what-have-you.

Nate specifically mentioned this as proprietary, thus we will likely never know. It’s not hard to imagine though - some combination of ramp test results, indoor/outdoor workouts, outdoor rides, etc, going back “X” number of days. Or forever.

It’s interesting to read these comments (the whole thread, not just one or two comments/posters) and realize how difficult it will be to get people out of their own limitations and into something new. I am not an early adopter, but to me this seems to be so obviously superior to other training platforms, including TR’s current. Probably quite obvious from my posts, but hey, I should probably mention it :sweat_smile:

3 Likes

Not really true. If you take the same amount of time in both work and rest and total time, they are.

What do you mean? I’m pretty sure Nate did say that.

It was 4x10 and 40 minutes that was mentioned.

If you are being selective, maybe, but that is not how you measure the intensity of the ‘focus of the session’, TR are most likely doing what some of us have been doing for years, looking at the work and the recovery ratio. I seen a few people of here explaining this by highlighting the intervals on the workout review / summary page.

40M @ 90% = 0.9 IF
4x 10M @90% r5 45% = 0.83 IF

To say you can’t see which one is harder (intense) using TSS and IF is misleading IMO.

Consider the duration of the whole session and you get the full picture.

1 Like

And…what happens if you keep the duration of work and rest the same in both examples…

2 Likes

But to my simple mind you are now comparing a 40 min wko with a 55 min? Throw in your 15 min z2 before or after the interval in example 1 and you get the same IF?

1 Like

Of course but in reality there is no recovery in the 1x 40m, who adds the recovery intervals to the end of a 1x 40? If you add the same TiZ / duration at each intensity you get the same TSS and IF, no one is debating that, I am saying you can use IF, as users of TR have done in the past to look at how hard a session is, there isnt any real debating that, it is a fact.
Is it easy and intuitive to use TSS and IF, No but you can. The scoring system makes it easy for all users to understand what the level of the workout is compared to others without having think about or understand TSS and IF. For the purpose of AT it is needed, but its not magic.

1 Like

Just trying to get him to see that the IF and TSS would be the same, so what Nate said on the podcast was true.

I thought

was a little misleading :slightly_smiling_face:

What me? misleading not really, clearly stated and clarified original post, because like I said you dont add the recovery valleys to the end of the session, in reality you add some Z2 or even Z3 which then gives you a higher TSS

1 Like

I’ve been thinking (uh oh). I would be really hesitant to adopt Adaptive Training until other rides/workouts can be accounted for. I really don’t think it should be publicly released until this is available.

I’ll get this out the way - I’m training for an Ironman. However, I’m not necessarily asking for the runs and swims to be supported off the bat. It’s a cycling platform that offers very basic multisport stuff as an aside. I know that, I can’t wait until it’s integrated more but until then I’ll keep this cycling focused.

Since it’s announcement, I’ve been more mindful of my training. My worry is the AT system is being beta tested with TR staff and, as they said, specific TR users “you probably know who you are”. Surely this is an oversight. Maybe this is still very early beta but the way I see it, to prove your concept you’re going to need a much more diverse beta group other than just hardcore TR users.

I point this out as I, like a lot of people, have been using TR for my structure but using zwift and going outside for races and greater interaction. Just last night I did half of my plan builder prescribed workout intervals before I jumped on to Zwift with my TTT team for a course recon and finished my workout there. Yes, not as structured but it was a similar effort. As I understand it I’ll need to justify why I stopped that workout early with the new platform and it will in someway effect future workouts until the zwift ride that I did straight after is accounted for. The same goes for the long rides outside. Am I hell doing 3+ hours on a turbo, just not going to happen.

As far as I understood the timeline, it’s closed Beta (full AT system), mostly for basic bug testing and other obvious issues. Then, once they grow more confidant about stability they keep adding more and more users to the beta (which should also be more diverse).
But for the public release, as I understood it anyway, it won’t be switched to full on AT overnight at some point. Instead they will release smaller subsets of features slowly one by one. Presumably the full system will be active behind the curtains, but I guess the effect on the user will be limited at first.

1 Like

That was my take as well.

If everything’s going well, but not to well :wink: the ‘standard’ plan won’t be modified by AT until something is detected that indicates a requirement for change. Which is probably less likely until more features are in place.

There are many here who only train indoors and on TR. :man_shrugging:t3:

They are sitting on over 100 million rides. I would argue that’s as diverse as it gets.

If you mark the right category it shouldn’t be an issue.

As a fellow triathlete I would argue you should do that. You won’t get more bang for your time which leaves you with more time for the swim and run.

4 Likes

This might be true but what percentage is useful for AT and more importantly how do you identify and create a subset of data that can realistically be used for AT? It must be pretty difficult or at least time consuming.

It also worth noting that pretty much none of this data has subjective feedback, that will start to be collected going forward. I would suggest this is true for a lot of ML features particularly the more advanced ones and it is going to be a fair while before there is a decent data set for these.

The majority for it will be useful and likely also used. In terms of setting it up you start with a small sample and go from there.

Subjective feedback is helpful though not required. They could simply pick a subset of athletes and track their career. The context provided will help to detect patterns. Like when athletes fail workouts and how they fail (as discussed in the podcast).

We did the same thing with our branch network. AI and ML ultimately enabled us to sharpen our product profile which lead to a significant increase in revenue and customer satisfaction. We did it without subjective feedback.

3 Likes

True. You’ve go to have decent amount of data to get those smaller sets.

My point is there is a lot of potential ML features they have no data for at all, they are going to have to find ways to collect it. I think it has been mentioned in a round about way during the youtube video and in various threads.

1 Like

100 Mio. should be enough to derive those subsets.

In terms of data sources you are right. Though the power and heart rate data along with workout compliance is already a lot. Sleep, HRV, RHR, steps, stress, fatigue, weather, weight, yada yada yada would also be super useful. I am sure they will find a way to consider that when moving forward. The subjective feedback could be a good proxy in the meantime.

1 Like

I think you’ve slightly missed my point here. I’m saying lots of people use the platform in different ways. If the closed beta group consists of just one type of TR user, that being the indoor and TR only user (I think it’s actually more diverse than this) then you’re only getting a narrow viewpoint as feedback. Hence the post ride survey in the announcement doesn’t have a “finished ride elsewhere” option (I know that’s a bit ridiculous but I think the points still stands).

I agree with @Pasque, however, the closed beta is probably for a more technical stand point bugs etc. and a more open diverse beta group will be included at some stage. And a staggered roll out of features makes sense. If you ride indoors on TR only then the new features might be useful to you sooner, but as I do not fit into that category I would need to wait until they are able to analyse all rides to feed in to the new system, otherwise I would be concerned that the AT system wouldn’t be learning correctly as my training data would have blind spots that it’s not taking into account.

To be fair to them they have said they are going to address this I’m just stating that I don’t think it’ll worthwhile for me and many others until they do.

I just don’t enjoy 3+ hours on a turbo so I’m not going to do it. I’d rather put a jacket on and ride for 4+ hours. Everyone’s different and that’s a good thing and exactly what AT is trying to address.

1 Like