My Form ( Under Training status & Form) has been in the Red = “Very Tired” since Oct 11th and the trend keeps going farther down! the Recommended Workout is “Recovery” workout. So this morning i attend to do that b/c i have been doing structured &Targeted indoors training for about 3 weeks, but after the warm up my legs felt light and fresh so I ended up doing 1hr Vo2Max ( 6X4min @ 330watts) 271 XEP and 99 XSS. not only that, i went and showered and i felt fresh again and I whipped another 1.5hr Tempo ride with a 4X 20Sec sprints @ 550 Watts. 221 XEP and 110 XSS. Now my Form went even farther down and i think i can do 1hr ride tomorrow of Tempo/Sweat Spot. My question is do I understand the Recommended Workouts correctly or there is something else am i missing?
Hi how much history do you have loaded?
The Exponentially Weighted Moving Averages used in the estimates are sensitive to not having enough data to fill the window. From Wikipedia article on EWMA: “Whatever is done for [the first observation] it assumes something about values prior to the available data and is necessarily in error. In view of this the early results should be regarded as unreliable until the iterations have had time to converge. This is sometimes called a ‘spin-up’ interval.”
BTW one of the cool things about EWMA is it incorporates the concept of process that responds to an impulse that decays over time which is one of the key idea underlying ideas behind the Bannister Model of Exercise applied here.
In the case of the XERT Training Load the window is 42 days which is a key factor in the estimates. This means it will take a bit of time until that estimate has “converged”. On Training Peaks (using the Coggan Model variant of the Bannister Mode l) you will read that the model has to “learn about you” and isn’t that accurate for about 1 month (if remember correctly).
Hope I got this right Armando
[Also if I understand what Armando is saying we will soon have something cooler than two fixed windows (acute/chronic) EWMA]
[Also the XERT power duration curves and fitness signature stuff is already going considerably beyond the Bannister model by allowing us to make an “instantaneous” estimate of current fitness modulo the usual measurement / modeling challenges – again according to my understanding]
I need to hire you George. 
Shar, looking at your data, your Training Status (which is a reflection of “spin-up” as George described) is likely being underestimated. Many times, you enter into a period of training with training load that isn’t captured in your power data. This means that your TL is actually higher than what the power data indicates. Since the recovery is based on your Form as a function of TL, if your TL is lower than it really should be, Xert will show you as “Tired” when you’re not. Once your TL moves up to where it really should be, the Form indications should be better.
Soon we’ll be adding the ability to seed your TL with a starting value and to allow you to include manual activities with XSS. This should enable to you have Xert track your Form more closely.
Hi Armando – happy to contribute! I see so much potential here. Let me know if I can be of some small help.
Thank you.
We are seeing growing interest in what we are doing. Those that “get it”, like you, are immediately drawn and spend time learning and understanding the new concepts and ideas. However, there are still many that don’t, mostly because they don’t have the time to invest in understanding all the new concepts. They are different than what they have already spent considerable time understanding. Getting the word out and helping people appreciate what is being provided will not only help us in attracting new users, but will also help others obtain more value and benefit when they use Xert.
We do what we can to help others understand but people like to hear from other users more than us. We are pushing hard to get media to post reviews of our software as that will encourage people to invest the time to “get it” too. Once our users start speaking up, especially those like yourself that can articulate ideas without bias, the more coverage we’ll get and ultimately the more benefit everyone will get.
So help us spread the word. Everyone wins.
Hi Armando –
Been thinking about this – (dangerous)
First I am not part of a big online or “real world” community that I can influence.
But in my day job I do “data science” (as if there was any science w/o data haha) and what seems to be working for me is to find ways to connect the complex machinery to immediate value (not easy of course). For XERT an easier path from a lot of complex numbers etc. to something us customers want would be helpful.
More direct path to value – some ideas:
- content created by a cycling coach(s) about how to apply XERT to user goals (both TP and TR have this)
- a user editable Wiki overseen by XERT that explains more about how XERT works, why its better and how to use it
- Dynamically adjusting 4-6 week training plans that are able to incorporate an unplanned ad hoc club ride into the training plan
- more online coaches
- relationship(s) with companies that provide virtual cycling experiences such as https://www.dcrainmaker.com/2016/11/trainer-apps-guide.html
Thanks again. Many of these are currently in the works. The Wiki page was an idea thrown around a long time ago and I’m glad you brought it up again as it is a great idea and one that we should initiate. Having our metrics and real-time data on screen in a virtual cycling experience, eventually moving to data for HUD devices seems like that very likely direction. The coaching and sports science community are starting taking note of Xert and becoming familiar with our unique abilities. We’ve been receiving more and more calls and requests for new features from coaches who see many opportunities to use Xert in their studios and with their athletes. Soon things will tip in our favour and if coaches aren’t talking or able to talk about and use our concepts, they may start to miss opportunities.
Sounds good! Of course I would be happy to contribute to the Wiki (under XERT team guidance). Good learning experience for me. Suspect the coach driven content (looks like you have a pro team trying out / using the platform?) seems like a very fruitful path as they are in the business in translating all this complex stuff to their clients and are looking for an edge over other approaches.
Hello George Collier, Thank you for your input and sorry for the late replay, I was waiting to add some activities before I write up anytime! with that, I have 44 activities since Sept 5th and 16 of them are after I wrote this on Nov 3ed and to my chagrin the trend is still the same regardless of what you had stated the need of 42 times of activities. Armando, I completely understand I don’t “Get it” but what the percentage of your audiences get it? how many of them are data scientist, engineers, mathematician…etc who can perceive or comprehend EWMA, Bannister Model or whatever out there on Wikipedia ? I mean really! most competitive cyclists are not academically smart, they spend most of their time in the bike. I think having more videos is the best way to learn about Xert. I mean why I have to learn from this gentlemen about how to use your product https://www.youtube.com/watch?v=7tbfbe_0D0Y and Here https://www.youtube.com/watch?v=P1u3oLroef4 when I think it is your responsibility to deliver it to your customers. you had a good run when you started with your videos 7-8 month ago and then you stooped :(, I learned a lot from them and they just resonate better than reading text. thank you.
Thanks Shar. We’ll get back to developing some new videos soon. You are right that there are parts of our software that may take a moment or two to think about before they are understood. There is a collective understanding and knowledge sharing that will eventually take hold. I think it is a good sign that there are others that are sharing how they are using and getting value from Xert.
You may consider setting the Seed value under the My Fitness/Advanced tab. This represents what your Training Load was at the time of your very first activity. It’s a bit of guess work, if the concept of Training Load is new. This will give you a better overall picture of your progression with the XPMC.
Thank you again. I looked to fill in the “seed” value. But again I didn’t know whete to look for these value shall along with TL Time Constant or RL Time Constant…etc ? I’m looking at my graph trying to find something I can take out!, I still do not “get it” bear with me.
I’m trying to figure out the patten/ trend for my TL, RL and I could not. Sometimes goes up in my REST day and other times down then I go and train very hard so it goes down then the up then down …very chaotic. I still don’t have enough activities for the model to generate the values?
I suggest leaving the time constants alone for the time being. We’ll get into making changes to them as we start adding some newer capabilities.
Figuring out the trends in TL and RL isn’t easy. It’s important to keep your data up-to-date. In many ways, our XPMC is like other tools out there at the moment and you can apply similar practices. Alex Simmons has a blog here that is a good reference point: http://alex-cycle.blogspot.ca/2013/03/a-time-for-bit-of-sensitivity-analysis.html . TL is like CTL, RL like ATL and TSB like Form.
That is one awesome video – always wondered about that. The “one size fits all approach” always seemed like a potential major flaw. I understand that Bannister actually kept refitting the constants … maybe that gold is buried in the fitness signatures?
The frequency at which we can establish fitness from regular activity data has enabled us to determine Banister’s model parameters for all three energy systems - peak/high/low. They each have different k constants and time constants. In fact, we are looking to establish k constants and time constants for sub-populations, and if testing goes well, provide a regression routine to obtain an athlete’s unique set of values. This will depend on having sufficient data.
Interestingly, the P0 values in the model identify the fitness of an athlete in an untrained state. These must be individually identified. The time constants reflect the trainability and recoverability of the athlete and the k constants reflect the responsiveness to training and recovery, all of which would likely vary to some degree with each athlete but should stay fairly static over a training period. Projected values will decrease in degree of confidence as a function of how far in the future we are looking to predict fitness.
How the constants themselves are affected by training is another set of di mensions. One could easily expect, for example, that with years of training, P0 values go up. In effect, they too would have k values and separate time constants, albeit much much longer.
Ultimately, the intent is to hide this complexity and provide our users information they each want to know: what will my fitness be in the future?