Thanks
So will the system take fatigue from running or strength work into account for cycling workouts?
1 Like
The redlight/greenlight system already does consider fatigue from other activities.
Running example, before:
And after:
Strength training example, two days later, before going to the gym:
And after:
5 Likes
Because it has seen how you and similar profiles have responded to training in the past. If you’ve dramatically changed your lifestyle vs past habits then it might not be accurate. But nothing would be.
1 Like
Can someone link the beta thread, if there is one?
Based on previous Betas I’ve participated in, the beta thread is only open to those folks accepted into the beta
4 Likes
Historically TR hasn’t worked super well for me, but this new model sounds fairly intriguing precisely because it doesn’t care about that other stuff. It seems logical to try to include as many factors as you can try to build the most complete picture possible, but in reality what that ends doing most of the time is adding noise and complexity to make things less reliable rather than more accurate.
You can’t simultaneously control for every single factor that can influence your performance on any given data like work stress, weather/heat/humidity, how well fueled you are, a few nights of bad sleep, etc. But the one thing you can always do is measure the actual power your body is able to put out, and while all these other factors are hugely important in influence that power number, ultimately that number on the power meter will be an accurate representation of the sum total of factors leading up to that particular performance. Track that over time, map it out in relationship to the type of work you’re doing, and the ML can be pretty freaking accurate when it comes to spotting patterns. Overall numbers go up with lots of volume? Add more volume. Does it go up if you do two weeks of VO2Max, but down if you do three? That’s a pattern can be identified.
So @Nate_Pearson , am I reading this wrong or is this new model a relative simplification over the previous one? Simplification is obviously not the right word here, but it does somehow seem more straightforward and less complicated/noisy based on what I’ve been reading.
2 Likes
I guess this is my point though, it doesn’t know my lifestyle or habits so how can it compared me to other users of similar ages?
The current AI didn’t really work for me (the old Chad traning plans were what I responded best to) but I wait and will try them
It doesn’t need to know your lifestyle and habits does it? It needs to know how you respond to a certain stimulus. If your response is muted because you drink a 5th of whiskey every night or if it is great because you get a message after every ride, the model just needs to find others that respond similarly due to lifestyle, habits, or genetics.
7 Likes
IMO Even if you only spent a few hours on the forum once a month that would be highly useful for feeling like you’re in touch with the user base. I think that wouldn’t be too much of a distraction from running the business. Likewise for the podcast.
It can a bit aggressive though. I did a 16 min walk and it triggered a yellow day for me.
3 Likes
As I’m on the older end, unfit range and my goal is basically to get fitter, maybe lose a few kg, and stave off death (my family has a history) - my main concern is that it doesn’t over-do it for me and I keep enjoying rather than starting to endure. I need consistency, not injury. I consistently go with Moderate training approach or Masters plans.
Can anything be said about changes in those areas?
Not a huge deal, you’ve got more important things to work on at the moment. 
But if you’re interested, looking at last week as an example I rode 11:34 vs the planned 10:30 with a TSS of 563 vs the planned 507. Clearly above but in my mind (and my legs) sustainable for sure. However, AT wanted to make every non-endurance workout on the calendar endurance and scale down the ones that were planned endurance, sometimes to a rest day. I skipped them and hit the workouts and marked the endurance ones as easy. I had (and still have) a string of yellow days but haven’t gotten a red yet.
It’d just be nice to know what it saw in that. Was a big week and I’ve only been back on TR for a couple months so perhaps it didn’t think I could handle that. Ideally it’d say…you’re pushing it, we think this is a bit much but if your legs feel good…then keep with the plan.
I feel if it can hit that this is getting to par with an actual coach, to have those convos about how you’re feeling and adjust, or not, based on that.
1 Like
Yes, there will be articles, and other resources that go further in depth, as well as a podcast episode.
9 Likes
Will any of those singular sources outline ‘all’ of the changes? Like I said, the flood of information between what’s actual information/changes vs “get ready for this change” and numerous changes, is a cluster of information for someone who doesn’t want to dedicate a ton of time filtering through all of it.
Will there be a single source of information, either a post, article, email, etc., that will capture all of it? Your reply indicates that to actually understand all the changes, they will be spread over numerous articles, podcast, and ‘other resources.’
If the expectation is that every TR user needs to follow all these different emails, posts, articles, and podcasts to understand the changes to their paid-training app, that’s not the best method for a roll out..
Every person is different, and will need to approach things in a way that best suits them. A masters plan with a moderate training approach is a good place to start.
Alternatively, if you are coming back to training after a long time off, you could use TrainNow to recommend endurance workouts, until you establish a habit and work your way up to a couple interval sessions a week and then start a plan. You can also track other activities you might do including strength training, hiking, swimming and more.
The soon to be released predictive features of TR AI may be most helpful for you, in predicting fatigue and helping you see where your training can take you. Many athletes find these features helpful and motivating.
4 Likes
Run are included in the model, and it makes sense to me that a run could make you faster on the bike, or make you more tired. I haven’t looked into it in detail.
I do know that when we added runs, the model accuracy improved.
4 Likes
As LacticacidMCB said, it’s also looking at your personal past performance and not just others just like you.
So your genetics / lifestyle are expressed in how you train.
And LacticacidMCB is also correct that if you started a drinking habit and not sleeping, the AI wouldn’t know that and would have to learn about your new lifestyle through you struggling with your workouts.
3 Likes
It does to an extent but it doesn’t do it as well as cycling with a power meter IMO. It’s a data thing.
3 Likes
You nailed it!!! You wrote that so well, I’m marking it a “solution”.
A lot of ML work is experimentation and we’ve tried a ton of different things to see if they improve performance.
For instance, we tried adding the motivation survey and it didn’t help outcomes. I’d like to try it again, as it might help outcomes in just a small case, and that would be a win. But overall, the needle didn’t move.
9 Likes