Banister Model - Banister Helper

[Rui_B]

In Mark's Video explaining the Banister Model, he talks about adjusting the Positive Decay  and the Negative Decay in Banister Helper to help explain/understand our current physiology. These adjustments alter the RMSE. Is the idea to make adjustments to lower RMSE within the 50-11 day boundaries?  Is this what Mark was referring to when he says "making things better or worse"? Which should be adjusted first? Are there guidelines/suggestions for useful experimentation with these setting.

Cheers,

Rui

[Ale Martinez]

El jueves, 22 de abril de 2021 a la(s) 16:57:53 UTC-3, Rui_B escribió:

In Mark's Video explaining the Banister Model, he talks about adjusting the Positive Decay  and the Negative Decay in Banister Helper to help explain/understand our current physiology. These adjustments alter the RMSE. Is the idea to make adjustments to lower RMSE within the 50-11 day boundaries?  Is this what Mark was referring to when he says "making things better or worse"? Which should be adjusted first? Are there guidelines/suggestions for useful experimentation with these setting.

To have an idea of the magnitudes, see Table 1 in this article for Positive Decay/Negative Decay values found in the literature on Banister model. Negative decay changes are likely to have more impact.

[Rui_B]

Thank you Ale,

Looks interesting. Looks like my w/e reading is now taken care of.

Cheers,

Rui

[Rui_B]

Thanks again Ale, 

That helped put things into perspective. (It was a lot of TrainingPeaks Speak though.)

I found Mark's blog post after:

https://markliversedge.blogspot.com/2019/01/implementing-banister-impulse-response.html

...  this helped tremendously — I now understand the need to experiment with the setting.

... will be super interesting when machine learning is implemented!

Cheers

Rui

[Ale Martinez]

El domingo, 25 de abril de 2021 a la(s) 18:20:16 UTC-3, Rui_B escribió:

Thanks again Ale, 

That helped put things into perspective. (It was a lot of TrainingPeaks Speak though.)

I found Mark's blog post after:

https://markliversedge.blogspot.com/2019/01/implementing-banister-impulse-response.html

...  this helped tremendously — I now understand the need to experiment with the setting.

I would like to add a point I consider relevant: to build a good Banister model you need to do some test regularly, not to cherry-pick or test only when you are fresh, it is not pleasant, but if you want the model to "learn" how fatigue differentially affects your performance, better to do some tests under fatigue.

[Rui_B]

Thx Ale,

Are you suggesting the tests as a means of adding "fuel" for the  MMP filter,  supplementing for when best efforts are in short supply during a quiet time?

R

[Ale Martinez]

No, my comment, and Banister model, are unrelated to MMP.

[Rui_B]

Yes, sorry. I meant the filter (Levenberg-Marquardt) for cherry- picking out the best  3-20min efforts...  would the tests be supplemental for quiet times? I read Banister's paper(s) where he requires frequent tests — I was under the impression that Mark's Levenberg-Marquardt implementation  would eliminate the need for tests... though, I'm interpreting your suggest as, do tests to make sure there is an adequate supply of recent  max efforts. I can see that.

[Ale Martinez]

El lunes, 26 de abril de 2021 a la(s) 16:32:42 UTC-3, Rui_B escribió:

Yes, sorry. I meant the filter (Levenberg-Marquardt) for cherry- picking out the best  3-20min efforts...  would the tests be supplemental for quiet times? I read Banister's paper(s) where he requires frequent tests — I was under the impression that Mark's Levenberg-Marquardt implementation  would eliminate the need for tests... though, I'm interpreting your suggest as, do tests to make sure there is an adequate supply of recent  max efforts. I can see that.

Perhaps an example is a better way to express what I am trying to say: suppose you do an intensified block of training where you overreach and you are fatigued enough your performance -measured by the criterion test of your choice- decrease.

If you don't test in that period, because it is unpleasant, or discard the test data, because they are sub-maximal, you are losing information very relevant for the Banister model to learn how training load affects your performance.

Just my 0.02 cents.

[Rui_B]

Ale thats gr8 - thank you — I understand.

r

[Mark Liversedge]

On Wednesday, 28 April 2021 at 15:25:12 UTC+1 Ale Martinez wrote:

El lunes, 26 de abril de 2021 a la(s) 16:32:42 UTC-3, Rui_B escribió:

Yes, sorry. I meant the filter (Levenberg-Marquardt) for cherry- picking out the best  3-20min efforts...  would the tests be supplemental for quiet times? I read Banister's paper(s) where he requires frequent tests — I was under the impression that Mark's Levenberg-Marquardt implementation  would eliminate the need for tests... though, I'm interpreting your suggest as, do tests to make sure there is an adequate supply of recent  max efforts. I can see that.

Perhaps an example is a better way to express what I am trying to say: suppose you do an intensified block of training where you overreach and you are fatigued enough your performance -measured by the criterion test of your choice- decrease.

If you don't test in that period, because it is unpleasant, or discard the test data, because they are sub-maximal, you are losing information very relevant for the Banister model to learn how training load affects your performance.

Just my 0.02 cents.

This is really important, the filter used looks for maximal performances in the time period, so the sub-maximal efforts are sub-maximal when compared to efforts around them.

In some ways it might be simpler to replace the current method (which tries to fit a hull) with a far simpler extract of weekly peaks.

Mark

[Rui_B]

Hello Mark, I initially started this post asking re RMSE - using it to optimise t1 & t2…

I’ve re-watched your Banister video again (multiple agains actually)

I was watching RMSE tick up and down as you increased and decreased the Positive & Negative decay numbers in the Banister Helper window.

I am wondering if a better fit (lower RMSE) means more accurate decay numbers? Are 30-50days of fitness maintenance normal across the entire population? — As an UCI M55-60 athlete, perpetually getting ready for world championships in IP on the track, & TT on the road (when ever they may occur), I was floored this winter when I took 15days off from intense work, and had to lower my CP from 298 to 247 for things to make sense when I returned to some harder work. Its been 3 months of hard/smart work since, and I’m still not back to where I left off in December.

… As there were no maximal efforts, the Banister Model certainly did predict CP to fade (though not as much as actually happened)… Having a good understanding of t1 & t2 would be very useful in planning my training.

Cheers,

Rui