actually

I was in contact with the developer and it seems they feel they have disclosed the methods they use on their web site. It might be worthwhile to take a look and see what they say on their page http://baronbiosys.com/support-learn/.

Um, no.

That's marketing fluff. Not a technical description of their model.

After reviewing it, I'm still none the wiser as to the type/format of the machine learning algorithms (if any) that go into their model.

---

Alan Couzens, M.Sc. (Sports Science)
Exercise Physiologist/Coach
Twitter: https://twitter.com/Alan_Couzens
Web: https://alancouzens.com

We have not disclosed the details of our methods but we've been very open on the principles on which they work. (I certainly wish this would qualify as "marketing" but it probably is too much information for the vast majority of people). It's not "AI" or "ML" in the sense of, say, an NN. We have disclosed that we're using a 3-tier Impulse Response Model that is driven by work-allocated strain scores. Strain is measured as a function of power and MPA. We make no claims on the accuracy of the prediction model and rely on our users to judge for themselves how well it works for them. The outcomes that the models are based on aren't FTP or our Fitness Signature in fact but on the prediction of moments of failure that the model discovers and then uses to predict into the future. The precision of this method and the frequency at which outcomes can be observed is what distinguishes Xert and drives the system. The principles aren't new although have never been used in the way we have to make predictions and depict them in the way we have. We do have a number of scientists that we are working with that are looking to find new applications and insight. Full proof scientific of all the principles on which Xert is based will be difficult as they suffer from the observer effect. Others are rather academic and simply need a researcher to take the lead in publishing. Nonetheless you can easily judge for yourself the predictive value of what it does and don't need to rely on any claims, reports or "marketing".

I do look forward to seeing more on your NN implementation. We are looking at NN in combination with our models to bring applications to a broader range of sports and activities.

Best,

---

Armando Mastracci, Founder of Xert, an advanced data analytics and training platform. Blog, Podcasts

Maybe this was answered above and I missed it:

Along with the stress inputs do you need to tell it what are the good outputs or do testing (race) every once in a while to let the network adjust the weights or will it just pick up on the better results?

Thanks Armando,

I appreciate your honesty (and support!)

FWIW, for others reading, I have no problem with 'if-then rules' systems so long as they're advertised as such and not as "A.I." If athletes know who is writing the rules for the system they're in a better place to judge the worth of the system, especially in the absence of a loss function that quantifies the true predictive accuracy/value of the system.

That said, using a model that quantifies its true predictive accuracy is, obviously, always the better way to go.

Best,

---

Alan Couzens, M.Sc. (Sports Science)
Exercise Physiologist/Coach
Twitter: https://twitter.com/Alan_Couzens
Web: https://alancouzens.com

I actually enjoy reading your posts here and on the wattage/cycling physio groups.

I get the feel you are excited to explain your product but some people interpret it as marketing, which I don't feel is the case.

Thank you for that. I never felt I was marketing either. Just keen to tell folks that might understand and appreciate it. We do have a lot of very supportive customers and others that share in that excitement too which is great.

---

Armando Mastracci, Founder of Xert, an advanced data analytics and training platform. Blog, Podcasts

There was a purge?

There was. Back in late October there were about a dozen prominent ST'ers that got a permanent invitation to not come back. 🙄

---

My YouTubes

Thanks Alan.

It's not a true "if-then rules" system per se but more of an analytical model applied as a control system. I.e. math for the most part with regression methods used for some components. The math was obtained serendipitously, something I can only claim to have discovered and not invented.

---

Armando Mastracci, Founder of Xert, an advanced data analytics and training platform. Blog, Podcasts

"an analytical model applied as a control system"

Just when I thought we were making progress in moving beyond the buzzwords and accurately defining things 😊

---

Alan Couzens, M.Sc. (Sports Science)
Exercise Physiologist/Coach
Twitter: https://twitter.com/Alan_Couzens
Web: https://alancouzens.com

If I had said an AI based cloud service, then maybe you would have had me.

Analytical model: MPA (and hence your fitness signature) is effectively calculated rather than established through a method that has been trained or analyzed from historical data sets. We calculate MPA for the ride that is being analyzed. Some have called it an "expert system". I not really sure what the nomenclature is for what we do. Mike Puchowitz was asking how many parameters in the model and I said there aren't any (other than your fitness signature). For our IR models, we have separate time constants for the 3 systems. These were trained from our data and the coefficients are for each athlete are obtained through regression when rides are analyzed. The time constants are user configurable.

How the system treats each incoming ride is like a control system (analyzing all the data is computationally very expensive at this time .. we keep trying though...) Your existing signature forms the basis of the analysis as the algorithm looks for maximal efforts. A new signature is obtained when maximal efforts are discovered and the new signature is recorded (what is termed a "breakthrough" in the system). This is passed on to the next ride's analysis. Sort of like a feedback loop in a control system is the only way I can describe it. When a ride doesn't have maximal efforts, its analysis is skipped. These are then plotted and progress can be tracked.

Hope this helps. Thanks for asking.

---

Armando Mastracci, Founder of Xert, an advanced data analytics and training platform. Blog, Podcasts

Thanks Armando,

Much clearer.

So it's essentially a 3x Banister model for different points on the power duration curve?

Are the time decay exponents also updated recursively for the individual athlete?

Thanks for the additional info!

---

Alan Couzens, M.Sc. (Sports Science)
Exercise Physiologist/Coach
Twitter: https://twitter.com/Alan_Couzens
Web: https://alancouzens.com

Not exactly. A power duration curve is a by-product of the analysis. Work (technically strain) is allocated towards each system that drives MPA changes. These systems correspond to each fitness signature parameter. Since each parameter (system) affects MPA at every point in a power file, each system contributes to the generation of power. This *strain* is accumulated for each activity (or any time period) and passed into the 3 IR models. We summarize these concepts into Focus Duration and Specificity Rating. These enable the accumulated strain/training loads to be interpreted using the same ratios needed to perform at a given power output, converted to a duration to normalize it. This makes it more actionable as you then have visibility to what the "focus" is of your training, conceptualized as a point on the PD curve. In essence, your training can be added up, workouts and group rides alike, and what and how much is being trained can be determined in actionable form.

We don't yet update the decay time constants, although for some athletes we would expect to see sufficient information in their data to establish them. It is something we had considered but with the introduction of Freshness Feedback, this becomes less important.

---

Armando Mastracci, Founder of Xert, an advanced data analytics and training platform. Blog, Podcasts

More proprietary gobbledygook. I'm out. Thanks for the chat.

---

Alan Couzens, M.Sc. (Sports Science)
Exercise Physiologist/Coach
Twitter: https://twitter.com/Alan_Couzens
Web: https://alancouzens.com

I know you can get an EF inside the analysis of a workout and that it can be calculated per segment as well as for the whole workout. But I am not sure I've seen an EF graph. I'd love an EF graph over the season.

Sorry 'bout that Alan. Probably got a bit too far out over my skiis. Focus Duration is quite abstract and is the hardest thing for folks to get a handle on. The concept isn't proprietary. It's just hard to conceptualize.

I can appreciate your position. Thanks for indulging me.

---

Armando Mastracci, Founder of Xert, an advanced data analytics and training platform. Blog, Podcasts

Is the raw data available? I'd like to try Google's TensorFlow with the data. I didn't see a link to the data in the blog post.

AC, if you want to keep the interest of readers, don't come off as a prick know-it-all. You are on the fast track to Hambini here.

Since the winter season helps to fulfill a break to it's end there is some time to read about AI and ML .. my interests go where language does not only describe but even explain maths with help of draws .. these layers are really clever stuff to develop concepts for calculation .. this morning the firs time since four weeks the streets are dry and temperature above zero .. time to feed the cells with fresh air .. oSo >>

---

*
___/\___/\___/\___ the s u r f b o a r d of the K u r p f a l z is the r o a d b i k e .. oSo >>

.. to create a layer responsive helps to get an assess for an appropriate outcome you'd like to have ..

---

*
___/\___/\___/\___ the s u r f b o a r d of the K u r p f a l z is the r o a d b i k e .. oSo >>

.. ah .. and thanks for the help .. the concept of a wish and to handle all it's outcomes step by step in real time during processing is the biggest task I can imagine ..

---

*
___/\___/\___/\___ the s u r f b o a r d of the K u r p f a l z is the r o a d b i k e .. oSo >>

This is what really interested me as well. Sounds from above as though it will adjust the weights (e.g. using HR to normalise) so that it gets more data points without having to test too often. That would be a big plus for me.

Also then takes me on to another tangent as to what other data points it can take as input. E.g. I know my HR varies quite a bit depending on temperature, is typically lower at any given power cycling indoors vs outdoors, and to a lesser extent I see some HR variation depending on time of day. My HR also varies quite a bit depending on how rested I am, when tapered I hit higher HRs at any given pace/power. Presumably a sophisticated enough NN could pick up all those data points (it's all in the Garmin files) to predict fitness/performance better. A NN could also maybe make better use of the manually recorded data points that are captured in TP such as RPE.

Other thought is that I'm finding HRV increasingly useful for tracking non-training stress (travel, poor sleep, sickness, bad day at work, etc). Started using the HRV4Training app back in November, still learning how best to use the data but already it's apparent that in my case at least non-training stress can make the PMC largely irrelevant. E.g. in November I was pretty busy with work, spent a lot of time in airports, planes, hotels and high pressure meetings. Second half of December and YTD on the other hand have been very quiet. I've been disciplined with training, diet and sleep throughout (TSS per week, weight and average sleep hours have stayed fairly consistent). So using PMC you'd predict similar performance now to back in November. However if I look at my Recovery Points in the HRV app, they've jumped from an average of 7.6 in November to 8.5 in the last 4 weeks. And that actually tracks far better with my training/testing numbers, which took a big dip in November but are now a bit above where I was at in September/October. Seems pretty obvious looking back that the HRV numbers picked up on the work/life stress, and that that stress had a significant impact on my performance. So I wonder if you could feed in HRV numbers to a NN model as well, and use those as a proxy for measuring non-training stress and the impact it has?

Probably letting my imagination run away with me a bit here, but I'm another one who has always found PMC to be a fairly blunt tool at best and would be very excited if something better came along.

Hi Alan, last year when I read this article I was very interested both as a triathlete and as a ML PhD student.

How did you acquire the data to use in these tests? Is it available to share? I would be glad to play a little with that to...

Keep up he good work!

Alan,

This is super cool! Nice.

I'm a data scientist and have used LSTMs to take past 3 race times to predict next race time, and find—particularly using 70.3s to predict 140.6 times—much precision.

Some thoughts:

• 1D convolutional input layers on workout time-series metrics (HR, pace, cadence)?
• Compressed representations of workouts for ingestion by performance prediction models?
• RNN or LSTMs for performance prediction?
• RNNs for generation of workouts with inputs of (the above features)?

Cheers, M.
PS While I love calling all of this AI, these NNs are just ML... (complex function approximators)