Adaptive Training Closed Beta Update

The team is still working through bugs on workouts pushed to Garmin and done outdoors not showing and PL progression – based on my experience last week and now again this week. Interestingly, my Mon and Wed outdoor workouts this week did provide PL increases, but now again today’s did not. Today I did use an alternate stretch threshold workout with a higher PL than in my plan target (3.0 vs 2.4 planned) as I completed a 2.4 Saturday that didn’t show anything when it uploaded and ran through AT.

On today’s 3.0 Threshold I believe I was within target range, but got the struggle survey. Indicated no struggle, rated the next survey very hard, but no increment in PL. My support team buddy:) has all the info.

Learned from support that the evaluation is binary – either you hit the workout or not. So all my hard work to blow out the top end of the target range on intervals apparently has no impact (that does seem different that the concept of a super pass discussed a while back. You will never get an increase in PL that takes you above the level of the workout you’re doing.

Others may know the info above, but I’m working to learn exactly how the process works.

Ugh. Maybe they should just get rid of the up/down adjustments of difficult while they are at it :slight_smile:
I hope this is a temporary limitation, as I think this has things way too tightly tied to there being a ‘workout’ involved. I think the main case where tying it to a workout is the struggle/fail case, where it is important to know what it was you struggled/failed at. If that workout was too hard, then easier workouts should be given in the category.
In the pass case, I don’t think the assigned workout really matters - your PL should be adjusted based on what you actually did. If I go out and do a ‘ride’ that is really 3x20 at power X, this should count the same is if I do a workout that 3x20 at power X.
I realize that tying it to workouts simplifies things a fair bit, but once outside rides are analyzed effectively then this really should be more flexible. That part isn’t there yet, but I really think in 1+ years things should no longer be so rigidly tied to the selected workout.

Is this restricted to Outside Workouts?

1 Like

Had my reservations but credit where credit is due, the approach with progression levels is working surprisingly well. It is simple yet elegant and shifts the singular (and obsessive) focus from FTP to the whole power curve. Also find that it leads to more variety in the workouts that I am doing, where in the past I limited myself to SSBLV 1&2. Great job!
So far I did not detect anything in the Beta that would require machine learning, but might be missing something.

2 Likes

I don’t know @mcneese.chad ; I am only doing outside workouts so didn’t ask.

I’m right with you on that (get credit where credit is due, yes?) but what I understood from the emails with support is you only get what the structured workout is designed to deliver. Would love to be wrong on that – or even better, have it work as we’re discussing on actual results.

2 Likes

Two weeks ago structured outside workouts had to be a TR workout exported to Garmin, in order to update Progression Levels.

Currently my Garmin is getting workouts from TrainingPeaks.

Today a have 1 hour tempo at 75-85% workout.

If Progression Levels still requires a TR workout executing on Garmin, that means I would need to pick a TR workout and send that to my Garmin. For example both of these have outside versions:

and based on the course I’d pick White +1. Also, all my levels are 1.0 and I am ignoring the Not Recommended.

Do you know if TR-structured-on-Garmin is still a requirement? Or should I roll the dice, do the TP-on-Garmin, and just deal with possible no progression updating?

At the very least: 1) assigning workout levels to the workout library was done via ML, but that’s more of a static prerequisite to AT rather than part of the adaption process; and 2) classifying workout success/failure may use ML also.

Whether the actual adaptation process uses it right now, I’m not sure - since it seems quite simple in terms of picking an alternate from the plan that matches your PLs. But there could be more to it than that, and even if there isn’t currently, there definitely could be in future.

Roll the dice and see what happens.

I assume that AT needs a structured workout in TR’s database to determine which energy systems are getting worked and to calculate the PL. I think it then compares your ride (power, HR, etc) to the planned workout to determine if you passed or struggled.

As long as you have linked your TP and TR accounts, then rides created in TP get imported into TR. I would think AT treats these like any other custom workout.

I’m making a lot of assumptions but this is how I’d set it up.

I’m going to see what happens when both TP and TR are pushing workouts to Garmin Connect. Hoping worlds don’t collide on my 530 head unit!

image

2 Likes

Think both points are unlikely.
Regarding 1), that would assume the TR team just created a whole bunch of workouts and then subsequently used ML to rate them… That would be very inefficient and unnecessary. If you start mapping you will find that there are strong linear correlations between the workout levels and simple metrics like IF and total kJ. No need for ML, fairly simple analytics suffice to get the levels. Or reverse, this can be programmed into a workout creator to help guide the process (which is what the TR likely did; analytics are different for different zones).
On point number 2), that is a pretty basic problem to solve analytically as well, especially as the consequences of an error are very limited. So no need to create a big data set including human judgement (think physicians rating a diagnostic image) to feed a ML optimization function.

I am using plan builder. I don’t ever see “No Adaptations Required”. I either see “Adaptations Pending” or nothing.

Is there a manual/guide for AT? I am not sure about how to rate the suggestions. Maybe someone can direct me in the right direction? Search engine doesnt seem to show anything.

Eg. I stopped early one ride as I ran out of time and needed to get ready for work. So far my “excuse” for ending early was “equipment issue”. Another ride, I stopped midway as I had to wake the kids up to get ready for school. Obviously the second part of the workout was easier as I had the unscheduled “break”. Any suggestions to maximise adaption calculations in these scenarios?

Cheers

Mine didn’t start at 1. Was your ramp test before the adaptive training beta?

That’s precisely what they did with their whole workout catalogue. Nate discussed it pretty in depth in the announcement podcast. I’d recommend flipping through that video for more info on the ML side of things as that’s the best resource we have on it at this time.

Do you mean the surveys? If you’re being given a pass survey, just answer it honestly. There is no set way to do it as the ML is comparing your responses to your responses. My 4 and your 4 can be different and that’s 100% OK.

As for the other surveys, there is an option called Time for when you have to cut short for stuff like what you described. It would show up the in survey you’d get, but you may have to scroll down to find it.

3 Likes

I’m in. Only my outdoor rides (synced from garmin) are not (yet) evaluated. I cant do them perfectly due to external factors like traffic/crossings and corners. So for now i’m adapting the plan myself :slight_smile:

Important safety tip!

I was thinking of these worlds colliding:

Ha!

Anywho, it was underwhelming.

  1. TR workout didn’t want to load on Garmin. Everything looked fine. Wasted a half hour. It finally loaded.

  2. Forgot how much I despise TR’s use of Step Power Average on the default Garmin workout screen. Its terrible user experience, anyone recall Alex on the podcast saying he can’t ride with lap power average being display? Thats the only power number you get because of TR’s design decision. And that is one major reason why I stopped using TR’s outside workout over a year ago. Imagine a TR app where you don’t have 1-sec power displayed (for example 177W), and just a tiny real-time graph of power and average lap power. Now imagine because of traffic the first 20 seconds of a zone2/zone3/SS interval is 0W and how you are now stuck looking at a really low average power number. Very poor user experience as compared to structured workouts from Garmin Mobile and TrainingPeaks.

  3. Ended the ride, it auto-matched, but no survey on TR mobile app. No survey on TR web. Had to load the completed workout on TR web (on my mobile phone) and click the 3 dots to update survey.

  4. No progression updates, and based on some posts in this threads I’m guessing TR viewed it as a fail and didn’t update levels.

  5. TR still doesn’t show target power on the completed workout.

A few screenshots of my outside workout between 75-85%, with a bunch of short interrupts due to Friday night traffic.

Original workout targets on TrainingPeaks:

On TR with Tempo highlighting:

Some short traffic interruptions.

Calendar view this morning:

Survey I dug up after poking around TR web interface last night:

Gave it a “I didn’t struggle with this workout” rating:

And after that screen rated it Moderate.

3 Likes

Options available on Connect IQ

I just created a screen with the various power levels I wanted - NP for the interval, current 5s average, etc.

Start the workout, hit the down button (in my case) and I’ve got the data I want. I do get “Power high/low” alerts from the default Garmin screen but I’ll ignore those early on in an interval for the reason you give (amongst others)

1 Like

I’m going back to TrainingPeaks workouts on Garmin.

1 Like