TrainingPeaks announcing adaptative training,

Training
61

Agree. it’s difficult balancing act. The problem with fixed plans is that a lot of people just do what the system is telling them. So they have a crappy nights sleep and boom at 6am - VO2 intervals. people will stick to the plan and potentially fail the session. I like the approach where there is a suggested plan but with other options available. You’re not likely any time soon to have a system that can completely prescribe you workouts, but i think that plans that adjust and are flexible taking into account things like missed workouts, additional workouts etc are helpful.

I think what would be more beneficial are training plans, dynamic workouts that adjust to your abilities across different points of the PDC rather than just FTP. even workouts that adjust on the fly compared to the effort you put in. % of FTP based workouts whilst effective are not optimum because this approach assumes all athletes capabilities above and below threshold are the same. This is not true. A time trialist and sprinter may have the same FTP but their ability to handle workouts around and above FTP could be dramatically different. As I said, you will get better doing workouts prescribed as a % of FTP but you potentially won’t maximise your potential training this way.

62

This is what Xert tries to do with their smart workouts - continuously adjust power levels based upon “Max Power Available” (modeled parameter).

Ideally this might be the nirvana, but I’d settle for the following:

  • Personalized CTL & ATL decay rates
  • Personalized sustainable weekly TSS ramp rate - that is, what is the weekly average TSS I can increase by that I can sustain for XX weeks / months / a full season? Is it 5? Is it 10? This might change depending upon what my current CTL is. For example, I might be able to increase by 10 TSS until my CTL gets to 70, then only by 5 TSS from CTL of 70 - 90, and then only by 3 TSS between 90 - 100. Where 100 is the max I can currently sustain / hit without severely crashing
63

maybe you could hit 100 without crashing if you did enough hours low-intensity, going back to the “not all TSS are created equal” idea.

From my point-of-view, having followed a more traditional “time-crunched base” in the past, my experience has been CTL ramps up to 70-90 range are dominated by tempo and sweet spot. Personally I don’t seem to have much issue with TrainerRoad’s 5 weeks on / 1 week recovery during this phase of training. Ramps are straightforward at this point (with enough sleep) due to sub FTP intensity dominating the workouts.

Things change during build, due to the change in intensity and personally I don’t find tracking CTL to be of much value at this point. If I start doing all out vo2 efforts, or try and do TR’s three vo2 efforts/week, looking at the past its clear I need to increase recovery by dropping to 2 weeks on and 1 week recovery.

Not sure I would find much value in AI-ML delivering personalized PMC stuff, but have an open mind about it.

64

I’m looking at customized CTL / ATL / weekly capability to increase TSS as the foundation for any type of customized workouts. That is, for an AI - ML system to be able to recommend a change in my planned workout schedule, it needs to have some type of model that says I can as is / I can handle more / cannot handle the current planned workout. And to my my way of thinking, it implies some type of model of what I’m capable of handling (and ideally the model would need a lot more: e.g., # of hours of sleep, weight, some overall “stress” measure, etc.). Hence why I think being able to accurately model these parameters is a prerequisite to AI - ML driven training.

65

Ah yes, welcome to the API access wars in the valley.

66

Here’s an interesting paper tangentially related to this topic. http://www.cse.chalmers.se/~jomoa/papers/MLCyclingWCPAS18.pdf

I have thought about developing something that automatically adjusts a workout in real time based on HR predictions using the methods described in this paper.

1 Like
67

Personally I find TSS to be a bad metric for tracking training stress. I can easily get 250 tss from an easy club ride with a cafe stop, but a hard cx race that leaves me with dead legs for days will only get me about 70 tss. I know from others that say tracking TSS/CTL doesn’t work if you race a lot.
I used to think it was due to TSS only taking the amount of work into account, and not the ‘quality’ (I know IF is in there), but now I think it is due to not following the individual variations above FTP, and not distinguishing between contributions from the different energy systems. I agree with you that TSS works pretty well at effort levels near or under FTP, but for me it is lacking at higher powers. You could explain that from the Coggan point of view - TSS only takes FTP into account, and not the individual variations that are expressed in iLevels. But even those I feel don’t appropriately take especually sprint efforts and other neuromuscular accelerations into account (its all just level 7), which I think significantly contribute to stress from races.

1 Like
68

This is exactly what Xert’s XSS system does.

Based on your Threshold Power, High Intensity Energy and Peak Power, XSS (TSS of the Xert world) is proportioned to each of Low XSS, High XSS and Peak XSS. This graphic shows the relationships:

XSS%20Work%20Allocations

These are then tracked in exactly the same way as TSS, as an exponentially weighted average. The default decay rates are as follows:

XSS%20Time%20Constrants
XSS%20Time%20Constrants890×136 7.77 KB

Training Load is equivalent to CTL and Recovery Load to ATL. Just as with the Coggan PMC, Form is Training Load - Recovery Load.

The clever bit is not just that there are different decay rates for each part that makes up XSS - you can see that the fatigue effects of efforts under threshold go away much quicker than those above, but the system can also look at the different parts in isolation to assess how fresh it thinks you feel.

Unfortunatly there’s no way to plot this information at the moment in Xert but it is posible to export all of the information as a CSV. I’ll plot it when I have time, partly because I want to understand the relationship between Form and the individual parts that make up the number.

Mike

2 Likes
69

I haven’t heard this before, do you mind expanding? I typically never go through speciality since during race season I workout with the team and have events on weekends. This year I’m starting early enough to sneak that block in. Is it the intensity leading up to specialty phase or is it the actual workouts during? How do you adjust? Replace rides or skip? Sorry I just am curious on what to expect

70

I think the workouts are too intense and are likely to deplete your ability to put out those hard efforts in races. Once you’ve got your FTP up to its peak with base and build, then for me the emphasis should be on maintaining form and staying fresh, particularly if you race regularly. For the most part, one hard session (a race, hill repeats, or chaingangs are good as they brush up race skills) and the rest of the week tempo at most should keep you sharp without losing your edge.

This article made a big change in my thinking:

4 Likes
72

I have now twice tried to follow the 40k TT speciality plan, twice it has left me feeling tired and with heavy legs, each time I’ve binned it and gone back to just doing my TTs for intensity and tempo/endurance rides with the odd hill repeat ride to keep VO2 at it’s pitiful level.

I’ve found trying to adjust the speciality phase around the UK’s time trial season just doesn’t work, you can have multiple races of varying distances in a week (although it’s more likely to be one, but if you choose you could race pretty much everyday), I’ve also found that as a time trialist the gap between my threshold and my VO2max isn’t as big as most, so testing and then trying to do workouts at a set percentage of one figure doesn’t work for me, I really struggle with the 120% of FTP workouts in TR’s plans, but can do longer efforts at a lower percentage as prescribed by WKO4’s ilevels or what Xert has me do.

73

Following up on my earlier post about how Xert handles fatigue, here are three graphs that show how the system works. This is my data for 2019.

XSS%20Fatigue
XSS%20Fatigue1685×867 95.9 KB

The first is daily XSS plotted showing the Low, High and Peak XSS components. Not particularly interesting on it’s own but it gives an idea of the proportion of the different parts.

The second shows overall Training Load (blue) and Form/Status (coloured dots on grey). The Training load line is accompanied by 3 scaled versions: +80% (green), +40% (orange) and -40% (red). These lines form the boundaries for overall freshness/tiredness which can be one off the below.

  • Detraining (brown): Form above 80% of Training Load.

  • Very Fresh (green): Form between 80% and 40% of Training Load

  • Fresh (blue): Form between 40% and 0% of Training Load

  • Tired (yellow): Form between 40% and -40% of Training Load

  • Very Tired (red): Form below -40% of Training Load

Note how the upper boundary of Tired is the same as Fresh, which brings us to the third graph: Form for High + Peak XSS.

This works in exactly the same way as the normal PMC but just looks at Training Load, Recovery Load and Form for High + Peak XSS. I’ve only plotted the Form as it’s all that’s used. Above zero in green and below in red.

High + Peak Form is used to qualify the Fresh and Tired statuses:

  • Fresh: High + Peak Form > 0

  • Tired: High + Peak Form < 0

What this means is that even although you could have an overall form that would put you in the bracket for Fresh (between 0 and 40% of Training Load), your High + Peak Form can pull this back down to Tired if you’ve done to much high intensity work recently.

The training advisor in Xert takes your status and recommends workouts on the following basis:

  • Detraining, Very Fresh and Fresh: Endurance and High Intensity workouts based on the phase of the training programme and the desired ramp rate.

  • Tired: Endurance workouts only

  • Very Tired: Recovery workouts or Rest only

I think this deals with the problem that a low XSS race or hard effort can leave you in fealing tired for days but a high XSS endurance ride can leave you fresh the next day.

You also can give manual feedback to the system on how you feel so that the advisor doesn’t give you something too hard to do if your status doesn’t match what’s been calculated.

I must admit that’s it’s only through doing these graphs, which you can’t plot derectly in Xert, that I’m really starting to understand what the system is doing. Seeing it visually is much easier to understand than blindly (to a certain extent) following the advisor. It’ll certainly help me whan I want to go off-piste a bit more.

Mike

8 Likes
74

I am also getting convinced that a more layered approach holds the key to the future of adaptive training, so thank you @themagicspanner for sharing your work.

On a separate note, I recall reading an interview with a researcher at Xert who declared how a deeper study of HR and HRV is warranted and holds the next key to unleashing even better response to our individual training condition (I have no idea if they are already building in HR feedback in their recommendations). Indeed, I am rather disbelieving that one could safely focus on “just power” or “just HR.”

After all, WHOOP measures several quantities through the 24 hours, and strain is computed by them regardless of power metrics (to which the WHOOP is completely blind, making its predictions sometimes remarkable, sometimes less so). In contrast, even if pulling HR data my understanding is that TR is still basing the current plans exclusively on power, and sometimes even with disregard for what HR and HRV may be telling us as individuals (and sorry, but if my heart rate goes through the roof I will pay attention, no matter what Chad is telling me to do!). So, just as TR has the omnipresent FTP to regulate its power levels, WHOOP likely uses average values of HR under strain to detect higher or lower intensity.

In short, it boggles the mind that two such radically (parallel? orthogonal?) approaches can be fielded, and each with respectable, though intrinsically limited success, when in all likelihood the correct approach should probably merge both power and heart in the model.

1 Like
75

Completely agree, HR+Power used together gives you so much better information than power alone. Aerobic decoupling for example, why blindly progress to the next workout if your body is showing signs of struggling with the current workload? This is where I think TR’s analytics just don’t cut it, especially in building an aerobic base, getting the foundation solid is key to be able to progress later on. No point in increasing the length of your sweetspot intervals if you are struggling aerobically, better to repeat the workout until you get through it with more control. The crazy thing for me is (and I might be wrong) but this can’t be that difficult to add into post workout data to give people more of an insight into what’s going on. It’s the small things like this that keeps me subscribing to TP.

5 Likes
76

Not disagreeing, but where is the study that shows that training to achieve better aerobic decoupling performance (i.e., lower decoupling value) leads to better performance (define performance = higher FTP)? Also you would need to control for confounding variables (e.g., temperature, hydration status, etc.) that effect HR, or you would be curtailing training based on “faulty” measurement of aerobic decoupling.

1 Like
77

Hi Mike,

It remains difficult to balance providing the background data and details of the methods that Xert uses and keeping things simple. This has been a challenge, made even more difficult because we’re basing things on entirely new concepts like MPA and multi-dimensional training load. Over time as both the awareness increases and our system evolves, we feel we have an approach that will resonate with a lot of athletes. Athletes like you and the help you provide are an essential part of this evolution. Thank you for your post!

2 Likes
78

I must admit that this is my second time around with with Xert. Previously I had used it in conjunction with TrainerRoad and WKO and although I found it interesting, I was little frustrated with the ‘black box’ approach and a slightly clunky interface. The fact that I wasn’t really following the training advisor didn’t help either.

This time I’ve gone full-in and plan to use it all the way through to my main event in April. Taking the time to not only understand what the system is trying to acheive, but also how it’s trying to do it was an eye opener for me. At this stage I don’t understand every aspect of it, but those bits that I have worked out make perfect sense.

I think this could help others get their heads around what the system is doing and as your interface improves I’m sure more of the data could be made availible to those who want to see it.

The new Podcast is also helping users understand what is going on. Long may that continue.

Mike

3 Likes
79

I wasn’t saying that solely training to achieve better aerobic decoupling leads to better performance, I was saying it is a factor in assessing whether or not your body is coping or struggling with the workload, and I certainly wouldn’t define better performance as anything purely FTP, other metrics such as fatigue resistance and time to exhaustion make a much stronger cyclist than a single number. My point was simply why not include it in the post workout analysis? I find the TR analysis quite frustrating, as although it is easy to use it only provides a very broad brush to look at a workout and your progression (or lack of).

80

Hal Higdon launched his adaptive run training app (in partnership with TrainingPeaks).

2 Likes
81

My point was where is the study that shows that better aerobic decoupling is correlated with better performance? I’m not saying it is or isn’t, I would just like to see some study on this.

I think of it like HRV or resting heart rate: these were thought to be correlated / predictive with recovery, but this hasn’t been shown.

So for aerobic decoupling:

  • How do you use this metric to show improvement?
  • Do you track the improvement in aerobic decoupling for the exact same ride (e.g., 2x20 at sweet spot)? For the same type of ride (e.g., all sweet spot rides, no matter the interval duration)? Or all rides, no matter the intensity?

Cheers,