🎉 🎉 🎉 Introducing Adaptive Training! 🎉 🎉 🎉

[Snowflake69]

[mattonabike]

@Nate_Pearson A few of us asked about this about a million posts up in this thread and I think it got lost. Are there any medium term plans to add prospective data into the featureset such as ‘on a diet’? I could see some of your menstrual cycle data being something like this as well. It is one thing to tell the model a workout went badly because you were 750cal per day restricted for 2 weeks (or a woman on her cycle at x point, or pregnant, or in a house that just had a new baby) and it is another thing to tell the model that before a workout and let it make better recomendations. The tech related to this feels like its in the same wheelhouse of work you already have in place.

[Nate_Pearson]

@BCrossen

Based on the podcast this sounds like what they have in place, and that they have the ability to run new features against the model to see if it further improves the accuracy. So there is some human intervention for hypothesis testing still, and I find that to be quite interesting too.

This is exactly right.

I like to think of ML like cooking.

Ingredients = data
Chef = ML Engineers

You need a great chef and great ingredients to produce something special. Just because someone else is doing “ML”, doesn’t mean they will come up with the same output.

Just like if I had great ingredients I wouldn’t come up with the same dish that Gordon Ramsey would.

[Nate_Pearson]

@Captain_Doughnutman

I’ve thought about this more and one missing link (there are several) is AT pulling from custom workouts instead of TR workouts.

I read an update today that this is production ready now. So most likely this will be in beta very very soon.

[Nate_Pearson]

@Captain_Doughnutman

Ok, I understand that part, but say I do a 2x60min SS session and don’t quite finish the last interval and report a high RPE…will AT assign a TR workout which will be less than optimal/relevant or will it suggest a custom workout from my personal WOC library?

If you’re on a plan, it will suggest a TR workout.

But…you could find a workout at the same level in your WOC and do that.

[Nate_Pearson]

@dbf

Ok, so Train Now is built on the AT data and workout rankings, right? So is Train Now a cheat for using AT until we’re invited in? I’m following a plan, and tomorrow, for example, I have Darwin (60 min. with 4x10-min. threshold) according to the standard plan progression. If instead tomorrow I use Train Now and choose 60-min. Climbing will I be offered what AT actually thinks is best for my current, albeit hidden from me, AT levels?

A little bit. The thing we aren’t showing you in TrainNow is if it is Achievable or Progressive. That data comes out soon.

So you might get a little bit harder workout or a little bit easier workout.

Once that gets released with levels you can probably do a bunch of AT stuff manually in your plan.

[Nate_Pearson]

@killroy123

The thought of being able to just wake up that day and go, “well I’ve got x time to train, lets see what options I have” and to have the program automatically suggest a few options is amazing.

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).

[Nate_Pearson]

@JAN_LESKO

And if my “pause” is meant to relieve myself?

If you relieve yourself in a “set”, it will probably mark it as a failure.

If you relieve yourself between sets, it will probably mark it as a pass.

IE if you’re in a 2x20, pee in between sets, not at 10 minutes into a 20 minute interval. If you did that, it changes the nature of the workout.

[Nate_Pearson]

@bobw

Features like Plan Builder and the Calendar underwent a similar genesis (sorry for the further unintended film related pun) so it does look like TR’s modus operandi: internal testing; limited user testing; larger beta programme; full release.

Exactly.

We’ve got a lot of eyes on us. There will be bugs in AT for sure.

The internet is not a nice place, and if we have a bug that does something silly, there’s a good chance that someone makes a youtube video about it and tries to attack AT.

There’s also internal load on our system (both on employees and the actual infrastructure). It can feel like a panic if you let in tons and tons of people and you have five mission critical bugs all come in at once.

[Nate_Pearson]

@hubcyclist

What’s clear is a lot of people are really fickle and demanding, and TR is often in a no-win situation where nothing is good enough. For a while people were annoyed at lack of masters plans, or annoyed with lack of options between low-mid-high volume, or whatever. Now that they get their next step revealed, people aren’t satisfied that it isn’t rolling out quickly enough to most users. There’s a guitar builder on youtube who says at the end of videos “if you think you can do better, build it yourself” I think the same needs to be said when people talk about training products

Yes…but…we can do better and we want to do better.

Please have your engineer friends apply at www.trainerroad.com/jobs.

There will also be a plan announcement tomorrow in regards to the volume steps between low-> mid and mid->high.

[mcneese.chad]

@Nate_Pearson

There will also be a plan announcement tomorrow in regards to the volume steps between low-> mid and mid->high.

Looks like this one might just get checked off :

• Feature Request: Training Plan +1, -1 Variations (Low-Mid, Mid-High)

[Nate_Pearson]

@gregkogut

And I select a workout that alternates between tempo and sprint efforts. Will it scale the tempo efforts and sprint efforts by two independent scaling factors so that the entire workout works as intended? Or does it just scale the entire workout by one scaling factor so that’s it’s achievable?

Right now, just the tempo part will be scaled.

But…the sprints in those are more like “spin ups” that real all out sprints. A real all out sprint would make the tempo part really hard.

[Nate_Pearson]

@GeorgeAnderson

I can’t understand why they do the last podcast and offer beta access to those that request it and don’t actually take anyone onboard aside from those already on it and those few that were already pre chosen. Its very disappointing. I expect we’ll wait weeks to get access.

The people we added didn’t have access before.

We had a few people in the summer get access too early, then we pulled that.

If you look at my post above, there’s one more issue to go before we let more people in.

[enki42]

@Nate_Pearson

It can feel like a panic if you let in tons and tons of people and you have five mission critical bugs all come in at once.

Panic? You mean it adds to the excitement. The panic is for the person who has to pay the bill to Azure when the developers spun up too many resources to handle the increase in load. You don’t want to pay for the 100 core box?

[Nate_Pearson]

@Flashpoint51

In the YouTube comments just before the podcast started, Nate himself said something along the lines of “if we announced we 'd found a cure for cancer, we’d still have 10% of users asking ‘but what about strokes?’”. I’m just glad they carry on nonetheless!

This is so true, but it’s a good thing! It just means we have an engaged audience who wants us to do better!

And we want to do it all!

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

[Nate_Pearson]

@roadbiketrevor

Interesting! Thank you for the info! This bolsters my hypothesis that if you hit 100% on almost all your workouts without a lot of skips/fails/turning down the intensity, the AT experience will actually be minimally different

Maybe. Also if your survey data is correct. There will probably be one person who is in the exact middle of the bell curve with no life changes and they won’t notice any difference at all.

In our experience, everyone gets lots of changes very quickly.

[Nate_Pearson]

@carbking

How will pee breaks factor into the adaptive training? Will their be an option to tell the system that I’m not tired, I just have a small bladder?

Yes, we trained the model with “pee breaks”.

Just don’t stop to pee in the middle of a set that would change the nature of a workout.

If you stop to pee in an endurance/tempo workout you should be good.

I might pee 2x during Baxter and it still says it’s a pass.

[kajet]

@hurdle

Multisport is really complicated to program for, let alone periodize.

Without question.
I’m not sore about the fact that TR is rolling adaptive training for cycling only.

I’m just noting that it’s pretty much a big meh for triathletes. Perhaps more importantly though, adaptive training has probably sucked up the company’s limited resources, preventing TR from implementing such long-awaited features as run training import. Again - just stating facts, aka speculation.

[ColoradoJames]

Sounds like I may need to invest in the TravelJohn/TravelJane company with all the pee talk.

[Nate_Pearson]

@enki42

For detecting overtraining can’t you see heart rate response during the training to the workout load? The way Garmin does at the start of workouts. Maybe keep recording heart rate after a workout to see it at rest?

It’s not that simple or standard across people. This is why we want the qualitative data to combine it with this sort of data.