🎉 🎉 🎉 Introducing Adaptive Training! 🎉 🎉 🎉

[enki42]

@Nate_Pearson

Please tell your friends to apply at www.trainerroad.com/jobs.

When there is only one type of engineer you’re currently looking for… But ok. Guess you’ll keep it on file

[PWeber]

Great - I do Phoenix variations regularly.

Will Tempo workouts ever be recommended in TrainNow under Endurance, if you have maxed out the Endurance Progressions for a timeframe or as a Progressive recommendation?

[killroy123]

@Nate_Pearson

This is the long term goal, but is not here today. We think we need more data about how you feel and why you stopped/failed workouts.

I suspect there will be an evolution of rules → ML in this regard. I’ll be sure to let you know when it’s a set of rules and when it’s pure ML (or most likely it will be a mix at first).

Fantastic. I don’t expect 100% tailored at the beginning but the idea of having a range of options suggested and then things tailored over time just sounds amazing.

Would I be wrong to assume that a future direction for TR is to sink more data into the program from other programs and devices to refine suggestions?

Love all of the work and super happy to see you and your team working hard to keep us happy.

[Nate_Pearson]

@patrickhill

Also, shorter V02 intervals seem like an entirely different energy system than long V02 intervals. I can complete just about any TR V02 workout at 90s or less, but I fail pretty much all of them at 3min or more. I’m just concerned that it will have trouble progressing someone like me. Any perspective here?

You are correct. In a plan, you’ll be in the same workout profile so you would stick with short shorts or traditional.

In Train Now, you might swap between the two.

This is one of the things the ML team is looking at. We suspect we’ll need finer grain control between workout profiles in energy systems.

We need a bit more data coming in to find out what the appropriate approach will be. I think this will be a learned thing to the individual.

This will probably only live in the backend and not be show to the user.

[Nate_Pearson]

@patrickhill

Of workouts commonly found in TR plans, which ones have the highest failure rates (I have my guesses here)?

The ones with the highest level have the highest failure rates.

To put it another way, the bigger the delta between your progression level and your workout, the more likely your failure rate is.

So if you’re a level 1 threshold, and you do a level 5 threshold, you have a higher failure rate than if you’re a level 8 threshold and do a level 9 threshold.

[mattonabike]

One more for you @Nate_Pearson. Can you share what non TR data skynet currently consumes without giving away the secret sauce. I dont generally feed garmin with weight data today, but if the model can use it its trivial to do so. If i wear a garmin watch to bed, does the machine see that data? Calorie data makes it from myfitnesspal into garmin if you want, will it get used? I know its a dangerous place to give away to much, so just looking to know what data many of us could easily make visible to TR that would actually be consumed.

[DaveWh]

@JoeX

I failed Redlake +4 last week, 5x6 at 108%

Which interval are you failing on? If the 4th or 5th, just drop the intensity by a few points on those intervals.

[enki42]

I’m not saying its that simple to detect but it is a data point that could be easily calculated and provided with no input from the user. So no burden on the user needing to do more (its hidden from them) and its a clean metric to feed into the model as cardiac decoupling calculations would be calculated the same for everyone vs just human feedback. I’m not saying to get rid of the other data points. Granted cardiac decoupling can mean other things too but that still is useful to track.

If the warmup and cooldowns are always the same, you could do HRV calculations based on a minute of the time during the warmup and cooldown. Seeing how you’re in the same position on the bike each time it should be consistent. Thinking the type of measurement hrv4training does. Obviously its not the same as right after waking up but this would be a good base that should always be the same exercise to exercise and between people. The reason for doing it at the start and end is the difference should show an objective measure of how hard was the workout. Not a complete measure you still need to ask the person but some insight into how hard your body thought it was.

[Jkauffman]

Nate, apologize if this has been asked.
I frequently do workouts when I am “on call” for work. As a result during a recovery period I will have to get off the bike for just a minute or two. So frequently a 4 min break will turn into a 5 min recovery ( i wont take the brake if the recovery is supposed to be less then a minute or two) . Also, when i backpeddle for a few moments during a recovery to catch my breath my trainer interprests that i have paused. Will these interruptions cause all of these workouts to be “failed”?? If so, i’m worried that my fitness will be underestimated.

[Nate_Pearson]

@Alen

Well all the features be available on the iPad/iPhone apps or is this something that is only on the Desktop app in the beginning?

Available on mobile/desktop/web…so everywhere.

@Alen

are the workouts weighted? For example if one was to do avalanche spire and have a super pass but following that have two days of rest versus someone who does eclipse +3 and the next day does polar bear and both workouts are just a pass are those weighted differently?

They are not weighted at the moment. We think we need more data about how you feel to do that sort of thing.

[Nate_Pearson]

@geebutbut

If a user is not able to continue building on their progressions (across the higher intensities) would AT be able to detect such fitness plateaus and determine that increases in overall volume may be necessary to continue progressing?

In our internal system, we sort of have that today. But, we want to make that for everyone.

The “sort of have that today” is that the FTP prediction will suggest different volumes for different people. And it’s not always a higher volume for people; sometimes it predicts bigger gains on a lower volume plan for some people.

The goal is to make everything dynamic and individual to the person. But that will take many small steps to get there. But what we have today in AT is like 10x better than anything else offered in my biased opinion.

[Nate_Pearson]

@JoeX

Successfully doing a workout with 2min vo2 intervals would place you lower on the vo2 scale than doing a workout with 4min vo2 intervals. I’m sure the ML does take into account any data it has access to but having two days of rest would change the fact that you had completed the workout and progressed or regressed.

It depends on the % of FTP, the length, the spacing, the recovery, the interval length, and some other stuff.

[Nate_Pearson]

@freoishome

Defining the goal is obviously paramount for Adaptive Training. In my case I have two goals, which may/may not be compatible, ie, road TT (20-40kms), secondly Olympic XC MTB. I’m sure I’m not the only one in such a situation. Are there more/less compatible goals, and hence determining the Adaptive training goal/plan?

This would be just like it is today with TR or a coach. You’d need to pick a primary goal and we’d train you towards that.

But those two aren’t that far apart. Especially if your XC courses have long climbs and you plan to ride them fairly smooth.

There’s no magic that I’m aware of there that can make you equally awesome at doing TTs and being punchy as compared to specializing in TTs or specializing in being punchy.

[Nate_Pearson]

@dbf

@patrickhill

I can complete just about any TR V02 workout at 90s or less, but I fail pretty much all of them at 3min or more. I’m just concerned that it will have trouble progressing someone like me. Any perspective here?

This Ditto. How will AT handle this?

Those workouts usually have a big level jump between them.

For example:
Baird +2 is a 6.6

Here’s Irvine that’s a level 10

If you pass Baird +2, Irvine would still be “Not Recommended”.

[Nate_Pearson]

@Buckethead

This is an important question that @zwillis1 brings up. You guys recommend doing additional z2 in addition to the LV plan quite a bit. Which I am sure many folks do. How will AT handle those 2 or 3 days extra days of z2 folks schedule? Should we look at using TrainNow on our z2 days instead of adding them to our calendar (I personally add z2 on days as a reoccurring workout)? Or will AT pick up the days that we do in addition to the interval days?

If it’s not on a plan, it won’t adapt at the moment. But you will get credit for those Z2 days. You probably want to make them “Achievable” though.

We’re going to have a LV+ and MV+ plan coming out soon. I’ll talk about it on tomorrow’s podcast.

[Nate_Pearson]

@CZS

Maybe I’ve missed this question in the gazillion posts on this thread, but for those of us who don’t use the Plan Builder, but instead put together a plan using the Base, Build, Specialty blocks will the ML and Adaptive Training features still work accordingly?

It doesn’t do that today but I believe it’s already built internally and we’ll have AT work on adhoc plans in the next couple weeks or so.

[Nate_Pearson]

@Mowag

This makes phrasing what I want to ask easier
If I have Leconte scheduled for tomorrow but add 3x15 threshold to todays workout (for the questions sake, don’t ask why I would do that :-)) → will I still have Leconte scheduled for tomorrow?

Yes, you can still fuck yourself by putting seven, level 10 workouts in a week and ignore what we tell you.

There’s still a portion of personal responsibility needed at this stage. But, if you follow the system I think it’s 10x better than before.

[Nate_Pearson]

@timb34

When we have AT, there’s going to be the Progressions graph. How is this going to change between different types of plan? Does the data behind it get tweaked for the targeted capacities so that you see all the bars increasing as you go through the plan, or will we see greater growth in the targeted zones?

You’ll see greater growth in targeted zones. IE if you’re a crit rider you’ll see more growth in Vo2 and Anaerobic compared to a triathlete.

@timb34

And what about training without a plan/event - does Train Now try to train you towards a balanced profile?

Train Now is meant for people without specific goals. Your profile will change based on the type of workout you pick. it doesn’t try to balance you out.

[Nate_Pearson]

@RobertSims

That’s a shame, was really hoping AT would work with separate TR plans rather than only via Plan Builder. I’m a duathelete and like the Sprint Tri speciality plan but prefer the General build plan rather than the Tri build plans…but no way I can force/get this via Plan Builder (…or is there?)

There is! Click on your build phase annotation on the calendar. You can select any other build you want.

[Nate_Pearson]

@mattonabike

@ambermalika Will we at some point be able to feed in higher tier information, specifically diet state? It seems like one thing to tell it ‘nutrition’ for a failure reason but it would be far more informative to the model to know “In caloric deficit/weight loss phase” where just holding steady while dropping weight would be considered progress.

Yes, we’ve thought about that. We’ve also thought about people having a “diet” option.

IE keto athletes will probably respond differently than high carb athletes.

This is just one of the many other data sources that we want to bring in.