mountain running with golden cheetah and power - graded pace

[Damien de Courten]

Hello guys,

several questions about golden cheetah related to mountain running.

1. I would like to model CP on altitude adjusted Power and not Power since I am training/competing quite high and without the corrections the data is much less interesting (can get very far from max when not adjusted). Why I can't fit the model on the critical mean maximum chart with aPower to get a CP_sea and?

2. More importantly is there any altitude adjusted W'_bal? Otherwise the analysis shows that I was all the time below my CP but it is not true once one looks at aPower.

2. what is the algorithm for measuring VAM? It looks a bit to low. Since it is the most relevant value (with eventually power) for us that would be interesting to measure the VAM for ascent only -as well as for charts

3. Any formula for normalized graded pace (training peak) or better Graded Adjusted Pace (Strava). It is not ideal but it gives a good idea for comparing workouts? I tried the following power expansion: 

SPEED*(1+0.0259210987316936*(SLOPE)^1+0.00181641090323033*(SLOPE)^2+-0.000100791815570037*(SLOPE)^3+2.24966715717189E-06*(SLOPE)^4+5.7138252399972E-07*(SLOPE)^5+-6.32478452806391E-09*(SLOPE)^6+-1.45448812276645E-09*(SLOPE)^7+7.14175600868895E-12*(SLOPE)^8+2.00322332001962E-12*(SLOPE)^9+-4.02676757059772E-15*(SLOPE)^10+-1.53051643192488E-15*(SLOPE)^11+1.06341531777731E-18*(SLOPE)^12+6.08885151706257E-19*(SLOPE)^13+-9.51860416228716E-23*(SLOPE)^14+-9.51860416228716E-23*(SLOPE)^14+-9.82832317286805E-23*(SLOPE)^15) This is based on the formula that Swiss topo use for determining the time required for each trail in Switzerland. It works for them, not for my purpose. Anything better? Thanks a lot! Damien

[Manuel Oberti]

Do you have a link please ?

THANKS

[Damien de Courten]

For the formula? Well it does NOT work for running. But here is the link to the empirical formula you can find in their excel table: https://www.randonner.ch/fr/signalisation/signalisation/temps-de-marche This si the formula used for all trail signalization in Switzerland  over a lot of km. Basically it is the time it takes for each trail in average. But I guess it works only for walking. It is very disappointing for a graded pace. It works much better on Strava.

Here you'll find a news article about it: http://user.math.uzh.ch/vorlesungen/sta120/fs18/web/5.Data/wandern_wie_lange_gehts_noch_tagesanzeiger.pdf

But it says that it does not work either for steep trails.

So I would like especially to know what is the formula for aPower on GC because this one seems to work for power and if anyone has a working algorithm similar to the one on strava.

[Ale Martinez]

El lunes, 2 de julio de 2018, 17:26:42 (UTC-3), Damien de Courten escribió:

So I would like especially to know what is the formula for aPower on GC

See https://github.com/GoldenCheetah/GoldenCheetah/blob/4022821b8296bfef9da202317bf3a6d2c9a21f7b/src/FileIO/RideFile.cpp#L2249 

[Damien de Courten]

Thanks a lot Ale for the Link!

Actually concerning the Graded adjusted pace the problem was not really my formula since it is very similar to the Strava Formula. I used the graphic they provided [2] by calibrating the graph and drawing a few points I fitted the following 8 degree polynomial formula:

Linear model Poly8:

     f(x) = p1*x^8 + p2*x^7 + p3*x^6 + p4*x^5 +

                    p5*x^4 + p6*x^3 + p7*x^2 + p8*x + p9

Coefficients (with 95% confidence bounds):

       p1 =        3968  (3092, 4843)

       p2 =        1264  (1096, 1432)

       p3 =      -935.3  (-1128, -743)

       p4 =      -290.8  (-323.8, -257.7)

       p5 =       21.71  (8.843, 34.57)

       p6 =       15.62  (13.75, 17.49)

       p7 =        17.5  (17.24, 17.76)

       p8 =       2.811  (2.782, 2.84)

       p9 =           1  (fixed at bound)

And it is not far from the Swissrando [1] formula. Here I compared it with the previous formula used by Strava they obtained from Minetti 93 [3]. I compared the three formulas and they are not far from each other. Indeed the Swissrando formula fails at large gradient [1]:

[1] https://www.randonner.ch/fr/signalisation/signalisation/temps-de-marche 

[2] https://medium.com/strava-engineering/an-improved-gap-model-8b07ae8886c3

[3] Minetti, A. E. et al. (2002). Energy cost of walking and running at extreme uphill and downhill slopes. Journal of Applied Physiology 93, 1039–1046.

[Damien de Courten]

So the problem for me to implement Graded Adjusted Pace as a User defined curve is actually not the formula itself but the slope data in the file. It seems it comes directly into the Fit file, am I right? It is actually always really noisy and looks really different than reality.

So is there any possibility to implement a moving average in a user defined curve on Golden Cheetah? I did not see anything about it anywhere.

On Monday, 2 July 2018 23:18:29 UTC+2, Ale Martinez wrote:

[Damien de Courten]

And is there any possibility to use aPower and an equivalent aW'_bal? for a performance and CP chart with curve fitting?

On Monday, 2 July 2018 23:18:29 UTC+2, Ale Martinez wrote:

[Ale Martinez]

El jueves, 5 de julio de 2018, 17:24:21 (UTC-3), Damien de Courten escribió:

So the problem for me to implement Graded Adjusted Pace as a User defined curve is actually not the formula itself but the slope data in the file. It seems it comes directly into the Fit file, am I right? It is actually always really noisy and looks really different than reality.

If the file has slope it is used, otherwise it is computed.

BTW, we already have the code for Strava GAP in Daniels Points, but it is not available as derived series or an independent metric, see https://github.com/GoldenCheetah/GoldenCheetah/pull/2776

So is there any possibility to implement a moving average in a user defined curve on Golden Cheetah? I did not see anything about it anywhere.

You can reference previous samples so it is possible, but there is no vector operations over series so it can become tedious if you want to average more than a couple of samples, in this case an EWMA would be easier.

[Ale Martinez]

El jueves, 5 de julio de 2018, 17:25:20 (UTC-3), Damien de Courten escribió:

And is there any possibility to use aPower and an equivalent aW'_bal? for a performance and CP chart with curve fitting?

You can do that in R/Python charts, using the standard CP chart a curve is fitted when the selected channel is apower but the parameters are now shown, IIRC.

[Ale Martinez]

Model parameters will be displayed in the next dev build when aPower or aPowerKg channels are selected, it was a simple change since curve fitting was already enabled.