Really want to start using adaptive training but until this is a thing I’m sticking to the old school way…
Nope, not yet.
Shame, it feels ages ago since this got teased… Has there been any talk on when this is ready?
It’s been mentioned on the podcast multiple times, even recently in the last 2 weeks or so. It is their top priority at the moment, but we have no stated timeline (which is the usual case).
Just curious, does anyone know why it seems to be so hard to account these unstructured rides? Can’t they just look at the power zones in the .fit-file, look how long you worked in each zone and update progression levels accordingly?
Easy in concept, tough in reality. Considering people are likely looking for “credit” to associated Progression Levels, this is far from simple, IMO.
Take your typical Tuesday Night worlds group ride. It likely consists of a HUGE range of efforts from easy spinning for a while, to insane kicks for town/sign sprints and whatever else between. Trying to “categorize” a random set of efforts and then adjust PL in an intelligent way that actually reflects the adaptations one could expect from those efforts is not a cut and paste operation.
Even people performing actual TR outside workouts end up with enough variability to call into question whether they got the intended demand and future benefits from the planned workouts.
Now take a totally random set of efforts, that likely are not restrained to a single targeted training zone/level like most TR workouts, and you get a hodgepodge of work that difficult to classify and “reward” to meaningful results for your PL’s.
It’s not just what you did but determining what was the intent. As @mcneese.chad notes, intersection sprints and the likes can just mess things up.
my current hill to die on is that I think the progression levels people think they achieve on unstructured outdoor rides will be a lot less than people think they are. I was looking at a spirited club ride and although it was a good tempo, it was maybe 10mins of solid threshold up a climb and a bunch of randomness that frankly wouldn’t really register on any progression level. I may be wrong, but I think folks may see this more often than not
screenshot of referenced ride to show what a mess this kind of stuff can be
I select an outside workout representing my outside ride, more or less. Place it in my calender, afterwards AT gives some sort of score. I do this at the end of the ride, with Strava on pause till the workout is added in the calender. Not ideal, but close enough for me. Better analysis is coming. Happy with Adaptive Training, so it will have to do.
Yup, I was about to share 3 different rides to show the variability. It is possible pull some zone averages and even find “intervals” as ICU and other apps, but really connecting that to the PL is not automatic like many people assume. There is value in the rides, but it is far from concrete as we get with a structured workout (inside or outside).
As you and others mention, I fear that expectations likely exceed the reality of what can be pulled from data and rides like this.
I suspect that rather than go the whole “let’s analyse this and try and guess what the rider was aiming to do” route that there’ll be a couple of survey questions:
“This looks like an unstructured outdoor ride, please select the closest option that describes the ride from the list below”
Then maybe a follow up question depending on the answer:
“Your Z2 ride was actually ridden close to your threshold, do you wish that the appropriate PLs are credited rather than the intended Z2?”
Exactly - the whole concept with the PLs is what were you planning vs what did you actually achieve, therefore what impact that has on your PL and whether future workouts should be made easier or harder on that basis.
We already see differences between the PLs on recreated custom workouts vs the exact same workouts in the TR catalogue - because of the way the PLs were developed by reviewing the catalogue of TR workouts and people’s performance on those workouts.
Nate’s red light/green light system based on recent load sounds cool though, and also sounded like it would be easier to implement since I don’t think it required PLs to be assigned to outdoor stuff (bit like how TrainNow picks workouts).
I am sure you’re right - I still think many (most?) people have no real understanding of how much time they spend coasting or in z1 and hence how little actual worthwhile sustained work gets done on these unstructured group rides. I do think the real benefit comes potentially from more structured outside efforts though, for example the long 60-100 min non-stop tempo interval rides I’ve had a lot of in my base plan.
Having said that, I think the value of understanding fatigue and how these unstructured rides might affect other sessions is potentially huge. AT being able to look at a Wed noght worlds or Sunday club ride, and decide what went on there means a different session might be in order the day after, would be a great benefit I think. Maybe thats linked with all the traffic light stuff being talked about, but I’ve given up reading about whats supposed to be coming any more, and will simply look at it when its launched
Super good points. I do hope that the detection aspect and even more intelligent plan adaptations are coming as well.
I’ve mused a ton on this problem and how I would handle it - modeling is one of my favorite things to do, and even more fun to think about in the context of cycling.
There are a few steps:
you need to disaggregate a noisy signal into different bins. Think of this as a decoding task of a signal into different categories.
you need to be able to classify (encode) that into a workout level for different categories - because unstructured rides often cross multiple ones.
you need to have an idea of cross-over effects between different zones —> this one is a bit of a challenge when trying to understand say endurance and tempo effects on threshold and VO2 efforts. Once you are in the upper sweetspot zone through threshold, I think the critical power/W’ model Skiba talks about likely does a good job of categorizing cross-over effects - this is especially nice because work-time relationship is strongly linear. I suspect a similar model/functionality could/will be made from the big dataset TR for all zones. it just may not look like a coherent model if it is a brute force ML analysis.
Related to the above, you need to understand the impact of both going into and out of a zone. Here there are hysteresis effects, where there is a recovery lag once you go out of a zone. That “recharge rate” is just as critical to categorizing workout impact as time in zone, and is harder to classify. This is a big problem with TSS for classifying a workout. TSS just aggregates and things get lost when crossing zones through work/recovery cycles.
If I were trying to do this, I would be looking hard at different power duration curve modeling approaches and using ML on my big data set to try and fill in the holes those types of models inherently have from being derived using much smaller, narrowly scoped datasets. I would also be using a sparse coding type approach for decoding/encoding my different power zone bins on any signal. Again, the key here is not just time in zone, but also the periods between time in zone.
A lot of these challenges go a way if you prescribe a certain work /rest period in a controlled way. TR has all the pieces necessary to handle the unstructured problem, but it isn’t trivial at all to do it. Certainly it is the most important missing piece they have right now to take AT to the next level.
I may be the typical weekend warrior, but I get over half of my weekly TSS from outdoor rides and perhaps my outdoor rides are too hard and generate too much fatigue for the adaptations they create, but I generally see about 150-175% of the TSS in the same time as a TR Tempo or Endurance workout would generate so even though there’s more faff involved, they’re just as or more efficient to generate TSS as indoor workouts.
The difference with my rides is I’m usually solo, so other than stops/starts at intersections given the relatively flat terrain the workouts are pretty consistent. I’d like to see something similar to Xert does that accounts for these rides from an overall training perspective.
I ride the LV plans on TR indoors and leave the weekend free to ride outdoors. I only ride indoors to get faster outdoors so would love to see this feature implemented soon.
For the time being I load up a workout I think best represents my outdoor ride and push it to my Wahoo. At the least it counts for something tho not fatigue but if it’s a SS workout for a Saturday it was already going to have a greater level of fatigue and I can always accept or deny any adaptations from it later.
I think this is especially good if going out and doing and zone 1 or 2 rides.
So, if you were to convert your MMP curve to work and plot it versus time, you would see it is probably pretty linear - at least to first order - for a pretty long time. So what that means is that the longer the duration, the more work you can do for that duration.
Bottom line is that you can generate a LOT of KJs in the tempo and endurance zone, and be primarily limited/affected by fueling there. That is going to translate over to racking up a lot TSS too, but it’s hard to quantify the net impact. TSS is a valuable metric, but it smears out a ton of relevant info for that.
I haven’t looked into it since adaptive training wasn’t available when I switched to inside riding but are TrainerRoad structured outdoor workouts working on wahoo now. I knew they were on Garmin and that they were having trouble with wahoo. Just curious if that has been sorted out.
Yes, TR Outside Workouts on Wahoo work well.