Losing faith in normalized power...



rmur17 said:
well I think that really only a TSS 'issue' and not on the NP topic originally raised. I don't see much point in further arguing about it here.
Em I still allowed to understand what additive versus non-additive means?
 
SolarEnergy said:
Em I still allowed to understand what additive versus non-additive means?
Sorry, no. ;)

He means that if you take the TSS of the work segments + the TSS of the rest segments, that sum will not equal the TSS of the entire ride as a whole.
 
frenchyge said:
Sorry, no. ;)

He means that if you take the TSS of the work segments + the TSS of the rest segments, that sum will not equal the TSS of the entire ride as a whole.
Yes .... and the example I posted ...

To further illustrate my point [note: I ignore 30-RA here] - look at a short 10-min warmup at say 0.5 IF versus 30-min @0.5IF. If one follows either warmup by say L4 1-hr at 1.0IF, the TSS scores are (IMHO) are both overstated - moreso the 30-min warmup.

The 'core' L4 workout is clearly 100 points. Short warmup: 100*10/60*(0.5)^2 = 4 pts. Long warmup: 100*30/60*(0.5)^2 = 12 pts.

Total workout TSS: short warmup should be 104, long warmup should be 112

But std. TSS yields 108 and 124 respectively or +4/+12 pts (unless I screwed up).


IMHO it would preferable that TSS add/sum/accumulate linearly per segment as I described in the 'manual' calculation. IOW, the whole should be the sum of it's parts and no more :)

But this is all pretty old stuff from Wattage ...
 
frenchyge said:
Sorry, no. ;)

He means that if you take the TSS of the work segments + the TSS of the rest segments, that sum will not equal the TSS of the entire ride as a whole.
Actually, this statement is true not just for work and rest segments but for any combination of segments in a ride file. The segments are not additive as separate entities.
 
acoggan said:
I'm not sure I fully understand what you're saying, but maybe this will help clarify things: TSS is non-additive even when calculated manually, e.g., using Excel.

Yea, I looked at some stuff and see your point. I hadn't tried with an example.

On the other hand, does it make sense to calculate an IF on a small segment?

If I use global IF (i.e. IF for the total ride) to calculate TSS for a single segment, the TSS is indeed additive (I did try with an example).
 
beerco said:
If I use global IF (i.e. IF for the total ride) to calculate TSS for a single segment, the TSS is indeed additive (I did try with an example).

Only at speeds << c. ;)
 
Of course! As your speed approaches c you'll become infinitely massive, or IOW, a track sprinter, and then TSS doesn't matter as your rides only last < 5 mins... :p
 
acoggan said:
Only at speeds << c. ;)
I understand the non-additive bit of the discussion ... but it did get me wondering (which I have been on and off since I read your and Hunter's book) about how you arrived at NP as being the 4th root of the average of the 4th powers of the ride segments ?

Not that I have any issues with it, I'm just curious as to the rationale/background for using the 4th power.
 
fastcat said:
I understand the non-additive bit of the discussion ... but it did get me wondering (which I have been on and off since I read your and Hunter's book) about how you arrived at NP as being the 4th root of the average of the 4th powers of the ride segments ?

Not that I have any issues with it, I'm just curious as to the rationale/background for using the 4th power.
Andy describes the derivation of the 4th power NP algorithm on page 9 of his original paper, "Training and Racing with a Power Meter." There used to be a link to the paper on the Midweek Club site, but I tested that link and it no longer works. I'll try to find a good link to that pdf file.

Here you go http://www.peakscoachinggroup.com/Power_Training_Chapter.pdf. But, on this version it's discussed on page 10.
 
acoggan said:
...which is interesting, since you appear to possess the physical characteristics (i.e., decent neuromuscular power and anaerobic capacity) that seem to favor production of "NP buster" rides.
Well I never go out to ride an "NP buster", just train and race. I always reckoned it was NMP that did the NP trick for me as my AC (IYRC from analysing my pursuit data) is/was a weakness - but we're working on that....
 
doctorSpoc said:
i think you may be missing JohnMeyers point.. he's saying that he feels these sub-Thresh NP variable intensity ride have over inflated NP values which get reflected in TSS which get reflected in ATL and CTL.. so if you are using the performance manager to gauge your training strain....
I don't see how this is a problem for the Performance Manager....

Let's for one moment assume that for these type of rides the NP/TSS are inflated over what they should be.

If these rides are a regular part of training, then CTL etc will simply reflect that - but CTL is all relative to your own data and it is the relative patterns that we look for. How are we travelling compared to last season's CTL ceiling, is my CTL ramp rate optimal, or what taper worked best for that event? Since these "inflated" TSS values are included in the data regularly (remember - they are a regular part of training), then you are comparing apples with apples anyway.

If these rides are not a part of regular training then the difference between 'perceived TSS' and 'recorded TSS' for a few rides would be very small in the big scheme of things and unlikely to generate a perceptable influence on ATL/CTL numbers.

So, it can only be a Performance Manager problem if the NP/TSS from these type of rides was being calculated inconsistently. I don't think anyone has suggested that.
 
Alex Simmons said:
I don't see how this is a problem for the Performance Manager....

Let's for one moment assume that for these type of rides the NP/TSS are inflated over what they should be.

If these rides are a regular part of training, then CTL etc will simply reflect that - but CTL is all relative to your own data and it is the relative patterns that we look for. How are we travelling compared to last season's CTL ceiling, is my CTL ramp rate optimal, or what taper worked best for that event? Since these "inflated" TSS values are included in the data regularly (remember - they are a regular part of training), then you are comparing apples with apples anyway.

If these rides are not a part of regular training then the difference between 'perceived TSS' and 'recorded TSS' for a few rides would be very small in the big scheme of things and unlikely to generate a perceptable influence on ATL/CTL numbers.

So, it can only be a Performance Manager problem if the NP/TSS from these type of rides was being calculated inconsistently. I don't think anyone has suggested that.
not going to spend too much time on this because as i stated in my post, i think that JohnMeyers' conclusion about NP being wildly inflated is wrong.. and you are probably right that in the larger scheme of things and using the performance manager to see relative trends (as i do myself and the way i think it should be used) even if it was off it wouldn't really make much difference...

BUT... if NP was really inflated, it's inflated.. i don't know how do you argue with that. if you are looking at your weekly schedule or a particular workout in isolation and trying to make sense out of it.. if this workout has an out of whack NP/TSS it could lead you to wrong conclusions or in performance manager and trying to figure out the why of why taper A worked but taper B didn't then this could lead to wrong conclusions... but i think the bottom line is that if NP were wrong it would be wrong... he was asking why can't he look at a workout or a week in isolation and get acurate data... why does he have to depend on smoothing and a happy coincedence of errors that happens to iron out another error... think that's what made him upset.. no one was addressing his arguement directly... just with indirect arguements to comments bordering on personal attacks instead of addressing his arguement head on.

his observations are actually just what one would expect and some were saying they are wrong. if they weren't right then interval training would be useless.. it's just his analyis and conclusions that were wrong
 
doctorSpoc said:
BUT... if NP was really inflated, it's inflated.. i don't know how do you argue with that....

think that's what made him upset.. no one was addressing his arguement directly... just with indirect arguements to comments bordering on personal attacks instead of addressing his arguement head on.
I'm not sure his argument could be addressed head on. His assertion for NP being inflated was based on RPE for a 3.5 hour ride, which included 73 minutes of off-the-bike rest, and almost 19% coasting, which he was then mentally comparing to the RPE of a hypothetical 2.25 hr steady ride at .86 IF. There never really was any argument against the validity of NP which could be objectively addressed.

doctorSpoc said:
his observations are actually just what one would expect and some were saying they are wrong. if they weren't right then interval training would be useless.. it's just his analyis and conclusions that were wrong
I pointed out to him in post #39 that his RPE was probably affected by all the rest periods, but that didn't go anywhere. Beyond that, the next best thing I could think of was to point out that even if he were correct (that NP was inflated) for *that ride*, it didn't really make much difference in the grand scheme of things.
 
RapDaddyo said:
Andy describes the derivation of the 4th power NP algorithm on page 9 of his original paper, "Training and Racing with a Power Meter." There used to be a link to the paper on the Midweek Club site, but I tested that link and it no longer works. I'll try to find a good link to that pdf file.

Here you go http://www.peakscoachinggroup.com/Power_Training_Chapter.pdf. But, on this version it's discussed on page 10.

This reminded me of a question I had from way back when.

Andy-- you mention in the Power Training Chapter that when you derived the formula for TSS and IF, the best fit for the power/lactate relationship was an exponential function, and that a power function using an exponent of 3.9 was nearly as good. You then stated that the exponent was rounded to 4 for simplicity's sake.

I've often wondered if you "un-simplified" the math and used a 3.9th order function rather than a 4th order function (or even the best-fit exponential function), if the NP/TSS algorithms would be even better and there would be fewer NP-busters. How much difference might this make?
 
frenchyge said:
I'm not sure his argument could be addressed head on.
i can see that you tried... but in post #34 he postulates, rightly that one of his assumptions may incorrect and he asks this..

"Correct me if I am wrong, but I interpreted normalized power to be an estimation of what power I would have to keep at a steady state to emulate the same percieved level of exertion." ...and no one corrected him, that would have likely went a long way in setting him on the right path..

but anyways...
 
Has anyone looked at that other file that goofy "JohnMeyers" fellow posted as further evidence to his point?

:p

Seriously though, this is all in good fun--a learning experience (hopefully) for everyone.
 
RapDaddyo said:
Andy describes the derivation of the 4th power NP algorithm on page 9 of his original paper, "Training and Racing with a Power Meter." There used to be a link to the paper on the Midweek Club site, but I tested that link and it no longer works. I'll try to find a good link to that pdf file.

Here you go http://www.peakscoachinggroup.com/Power_Training_Chapter.pdf. But, on this version it's discussed on page 10.
Thanks - I had read most of that paper before - perhaps I 'speed read' that section previously !
 
fastcat said:
I understand the non-additive bit of the discussion ... but it did get me wondering (which I have been on and off since I read your and Hunter's book) about how you arrived at NP as being the 4th root of the average of the 4th powers of the ride segments ?

Not that I have any issues with it, I'm just curious as to the rationale/background for using the 4th power.

Here's the way I think of it: TSS isn't "additive" over segments because the total NP is not equivalent to the average of the NPs of the segments.

This is because taking the "4th root of the average of the sum of the powers^4" over the whole range is NOT equivalent to the average of the sum of any subsets of "4th root of the average of the sum of the powers^4". For example:

(( a^4 + b^4 + c^4 + d^4)/4 )^0.25 does NOT = (((a^4 + b^4)/2)^0.25 + ((c^4 + d^4)/2)^0.25)/2

Since TSS is calculated from the NP, FTP and time....you can see that the only way TSS could be "additive" is if the NP of the whole ride was equal to the average of the NP of the segments. Make any sense?
 
JohnMeyers said:
Has anyone looked at that other file that goofy "JohnMeyers" fellow posted as further evidence to his point?

:p

Seriously though, this is all in good fun--a learning experience (hopefully) for everyone.
Mhm, I think Alex's and AC's advice/insight is the most valueable in this case, do the WO's in this manner and see if it's possible to achieve a higher CTL than with more normal WO's. If not, then the TSS ain't to high, if possible, then there might be something to investigate.
 
Tom Fort said:
This reminded me of a question I had from way back when.

Andy-- you mention in the Power Training Chapter that when you derived the formula for TSS and IF, the best fit for the power/lactate relationship was an exponential function, and that a power function using an exponent of 3.9 was nearly as good. You then stated that the exponent was rounded to 4 for simplicity's sake.

I've often wondered if you "un-simplified" the math and used a 3.9th order function rather than a 4th order function (or even the best-fit exponential function), if the NP/TSS algorithms would be even better and there would be fewer NP-busters. How much difference might this make?

As it turns out, very little (as I discussed at ACSM). Like any good modeler, I ran some formal sensitivity analyses on some representative files, and even if you, e.g., decrease the exponent to 2 or raise it to 6 it changes the normalized power by <10%. Narrowing or widening the smoothing window by the same relative amount (i.e., reducing it to 15 s or increasing it to 45 s) has about the same impact.