Calendar
Retailers
Stack & Reach
Custom Geometry Calculator
Online Training Log
MAIN
INDEX
RULES &
LEGEND
LOG
IN
Search
this forum
this category
all forums
for
All words
Any words
Whole Phrase
(
options
)
Newsletter Signup
Slowtwitch Forums
:
Triathlon Forum
:
Impulse Response Model is Dead, Eppur si muove ?
Tri Forum
Classifieds
Lavender Room
Jobs
The Womens
Print Thread
Ale Martinez
Aug 5, 08 17:10
Post #1 of 14 (542 views)
Impulse Response Model is Dead, Eppur si muove ?
Can't Post
Warning: This is yet another anecdotal evidence n=1 story, read at your own risk :-)
Objective: Try to fit real world running race results (Track/Cross/Road) with training data using the impulse-response model, in the line of
Modeling human performance in running
.
Subject: runner in his 2nd seasson ending with 4:30 1500 / 9:54 3000 / 17:15 5000
Training data: quantified using a segment by segment variation of TRIMP calculated from paces, original TRIMP formula with (HRavg-HRres)/(HRmax-HRres) replaced by Average Segment Velocity / Velocity @ VDOT.
Performance data: VDOT for Track/Cross and Road Races (VDOT and Velocity at VDOT calculated using J.Daniels Formulas).
Method: training / performance data was fitted to IRM model using
The RaceDay Performance Predictor
software.
Results:
Considering races from 5000 to 10k in the season (11 races):
Considering all races in the season from 1500 to 20k (17 races)
Full story (sorry, it's in spanish):
Modelo Impulso Respuesta
Ale Martinez
(This post was
edited
by Ale Martinez on Aug 5, 08 19:21)
Philbert
Aug 20, 08 9:41
Post #2 of 14 (474 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Ale Martinez]
[
In reply to
]
Can't Post
Ale's thread didn't get much response...but I think his data is pretty cool, especially since he used RaceDay to make the calculations and graphs, so I am giving him a bump. :-)
My Spanish is a bit rusty, however, there is a pretty cool feature on Google that allows you to translate a webpage.
Click here for the translation of Ale's page.
Of course, the downside of this is that I'm getting tech support calls in Spanish ;-)
Phil
Dr. Philip Skiba, PhysFarm Training Systems
Training Software for Triathletes. Training Advice for Champions.
Now available:
The Triathlete's Guide To Training With Power.
Andrew Coggan
Aug 20, 08 9:57
Post #3 of 14 (450 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Philbert]
[
In reply to
]
Can't Post
In Reply To
Ale's thread didn't get much response...but I think his data is pretty cool
Probably because the results just recapitulate what has already been shown in previous studies in the literature? (Although the pace-based modification of TRIMP is interesting...oh, wait: that's because it is the running equivalent of TSS! <g>) Still, I agree it is cool to see yet more evidence that racing (in this case, running) can indeed be viewed as a math problem! ;-)
BTW, what are the S.E.E.s of the parameter estimates?
Philbert
Aug 20, 08 10:38
Post #4 of 14 (407 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Andrew Coggan]
[
In reply to
]
Can't Post
Quote
Probably because the results just recapitulate what has already been shown in previous studies in the literature?
No, wise guy. I'm pretty sure most folks here don't have a subscription to JAP ;-)
Quote
BTW, what are the S.E.E.s of the parameter estimates?
We'll have to have Ale tell us for sure. I don't have his exact data set in front of me, but eyeballing the data I have against his .JPEG and pulling out everything except the race results, I calculate SEE = 13 watts. ( calculated as: sqrt (sum the squares of the differences between prediction and measured / number of observations))
Phil
Dr. Philip Skiba, PhysFarm Training Systems
Training Software for Triathletes. Training Advice for Champions.
Now available:
The Triathlete's Guide To Training With Power.
Andrew Coggan
Aug 20, 08 10:44
Post #5 of 14 (400 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Philbert]
[
In reply to
]
Can't Post
In Reply To
Quote
BTW, what are the S.E.E.s of the parameter estimates?
We'll have to have Ale tell us for sure. I don't have his exact data set in front of me, but eyeballing the data I have against his .JPEG and pulling out everything except the race results, I calculate SEE = 13 watts. ( calculated as: sqrt (sum the squares of the differences between prediction and measured / number of observations))
Phil
Sorry, I meant the S.E.E. (or CVs) of the parameter estimates themselves, i.e., the two weighting factors and their respective time constants. That's what you really need to know to understand the accuracy of the model/adequacy of the dataset.
Kevin in MD
Aug 20, 08 12:11
Post #6 of 14 (361 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Ale Martinez]
[
In reply to
]
Can't Post
Hi Ale,
Regarding your blog post as it reads from the Google translation. You mention that an exponentially weighted moving average is an approximation of the terms used in the Banister model. In fact they are exactly the same thing.
FITNESS(t) = SUM over all t's of k1 * [W(t) * exp(-t/s1] (Eq 1)
where today is considered t = 0
is exactly the same as
FITNESS(t) = k1 * {[W(t) / exp(1/s1)] + FITNESS(t-1)} (Eq 2)
Now, that said, I don't know if the cyclingpeaks calculates the EWMA for CTL that way, I don't know what they use for the exponential decay factor, I would assume that equation 2 is how they do it but I am not sure.
A minor point I reckon, but one that really speeds up the time it takes for your computer to do the math!
Ale Martinez
Aug 20, 08 16:59
Post #7 of 14 (300 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Andrew Coggan]
[
In reply to
]
Can't Post
In Reply To
(Although the pace-based modification of TRIMP is interesting...oh, wait: that's because it is the running equivalent of TSS! <g>)
Nope... It would be the running equivalent of Paulo's P_TRIMP, TSS is dead, isn't it ? :-)))
Actually we didn't use rTss or RunScore because they were not available when we start to collect data more than a year ago, and we don't use GOVSS because GPS data is a PITA in the track, splits for known distances are way better.
In Reply To
BTW, what are the S.E.E.s of the parameter estimates?
I don't know, if you like to play with the data I can send you the series (training and performance).
Ale Martinez
Ale Martinez
Aug 20, 08 17:12
Post #8 of 14 (287 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Kevin in MD]
[
In reply to
]
Can't Post
In Reply To
Hi Ale,
Regarding your blog post as it reads from the Google translation. You mention that an exponentially weighted moving average is an approximation of the terms used in the Banister model. In fact they are exactly the same thing.
FITNESS(t) = SUM over all t's of k1 * [W(t) * exp(-t/s1] (Eq 1)
where today is considered t = 0
is exactly the same as
FITNESS(t) = k1 * {[W(t) / exp(1/s1)] + FITNESS(t-1)} (Eq 2)
Now, that said, I don't know if the cyclingpeaks calculates the EWMA for CTL that way, I don't know what they use for the exponential decay factor, I would assume that equation 2 is how they do it but I am not sure.
A minor point I reckon, but one that really speeds up the time it takes for your computer to do the math!
Kevin, I think the recursive equivalent to (Eq 1) would be:
FITNESS(t) = k1 * {W(t) + [FITNESS(t-1)* exp(1/s1)]} (Eq 3)
in fact (Eq 3) is used in
Modeling human performance in running
I understand the EWMA in PMC is a little different, It's explained in Dr. Coggan's:
The scientific inspiration for the Performance Manager
Ale Martinez
Philbert
Aug 20, 08 18:00
Post #9 of 14 (261 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Ale Martinez]
[
In reply to
]
Can't Post
Andy and I discussed the SEE thing offline. It requires matrix math. I'm going to pass on that for now :-)
Phil
Dr. Philip Skiba, PhysFarm Training Systems
Training Software for Triathletes. Training Advice for Champions.
Now available:
The Triathlete's Guide To Training With Power.
Paulo Sousa
Aug 20, 08 19:12
Post #10 of 14 (241 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Philbert]
[
In reply to
]
Can't Post
In Reply To
Andy and I discussed the SEE thing offline. It requires matrix math. I'm going to pass on that for now :-)
Phil
That's some advanced math there, Dr Phil... You might have to hire a 1st year engineering undergrad to help you out.
-
pstriathlon.com
Philbert
Aug 20, 08 19:32
Post #11 of 14 (220 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Paulo Sousa]
[
In reply to
]
Can't Post
Actually, I was thinking of asking a know-it-all coach / engineer who has recently demonstrated his rather limited knowledge of physiology. But then I realized he'd probably rather just post occasional snide comments, because that's apparently the sort of thing that props up his ego.
Dr. Philip Skiba, PhysFarm Training Systems
Training Software for Triathletes. Training Advice for Champions.
Now available:
The Triathlete's Guide To Training With Power.
(This post was
edited
by Philbert on Aug 20, 08 19:39)
Ale Martinez
Aug 21, 08 14:48
Post #12 of 14 (149 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Philbert]
[
In reply to
]
Can't Post
In Reply To
Andy and I discussed the SEE thing offline. It requires matrix math. I'm going to pass on that for now :-)
Phil
Phil, thanks for your answer and support, my objective was to simply to apply the model in a real training/racing context for a track/cross/road runner to see what happens and share the results for discussion (not get published) and your software allows me to do this quite easily!
Performance metric used isn't perfect as we tried to normalize for different race distance/duratios using Daniels VDOT, not a fixed duration test.
Quantification of training load used also has his own problems, in particular it doesn't account for VO2 kinetics when speed changes.
It was interesting for me to see that, even with these limitations, the model seems to fit nicely!
My stats knowledge is pretty basic but I *think* the number of performance data I've used may be not enough to have reliable estimates for the 4 parameters.
BTW I've re-read Morton, Fitz-Clarke and Banister article (Modeling Human Performance in Running) and they didn't provide parameter SEEs, only R2 and F statistics for least-squares regression.
Thanks for all the responses and sorry for my poor english :-(
Ale Martinez
Philbert
Aug 21, 08 15:05
Post #13 of 14 (136 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Ale Martinez]
[
In reply to
]
Can't Post
Hey Ale,
Quote
Quantification of training load used also has his own problems, in particular it doesn't account for VO2 kinetics when speed changes.
If you want to send me the raw data, I'll enter it using RunScore and see if it makes any difference.
Quote
It was interesting for me to see that, even with these limitations, the model seems to fit nicely!
It is interesting. What is more interesting is that if you quit doing the regression after 8+ performance tests, and just enter training data, you find that the model predicts pretty well in the future. I'm working on a paper on that now that I finally have the last one in review.
Quote
Thanks for all the responses and sorry for my poor english :-(
Your english is better than our spanish!
Phil
Dr. Philip Skiba, PhysFarm Training Systems
Training Software for Triathletes. Training Advice for Champions.
Now available:
The Triathlete's Guide To Training With Power.
ACTN3
Aug 21, 08 16:03
Post #14 of 14 (114 views)
Re: Impulse Response Model is Dead, Eppur si muove ? [Paulo Sousa]
[
In reply to
]
Can't Post
LOL, so I guess you don't post on ST unless you can't help it ;)
"Consistency is the last refuge of the unimaginative." Oscar Wilde
(This post was
edited
by ACTN3 on Aug 21, 08 16:03)
Print Thread
Off Season
What will you do in the off season?
Bike focus
MTB or 'cross
No training
Run focus
Swim focus
What offseason?
XC ski or snowshoe
MAIN
RULES &
LOG
