BikeTechReview.com

  • Increase font size
  • Default font size
  • Decrease font size
Home Forum
Welcome, Guest
Username Password: Remember me

Rick Murphy's Normanalized Power Thread
(1 viewing) (1) Guest

TOPIC: Rick Murphy's Normanalized Power Thread

Re: Rick Murphy's Normanalized Power Thread 8 years, 6 months ago #24390

  • Ron Ruff
  • OFFLINE
  • Platinum Boarder
  • Posts: 894
  • Karma: 0
Kirk wrote:
In all fairness since he has the perspective of riding the course as a TT a bunch of times, it could also be that the group ride was an underperformance for that course for most of same reasons stated above. He did the solo rides as TT's too, so they are likely to be close to maximal efforts for that course. IME, group rides/races are almost never maximal efforts for the duration.


You are right... I finished "fresher" during the group ride than I do when I TT the course. It ends with a 1.5 mile climb at ~8%, and I managed to ride the best rider off my wheel, and left the other guy way behind. Both of these guys are stronger than me at climbing (and everything else) and one of them especially so. Even though the power fluctuations were high throughout the ride, I was conserving energy throughout (wheel sucking whenever possible). It wasn't my best ever effort up the hill, but it was better than I've ever been able to do it when I TT the course... and when I do that I always go up that hill as hard as possible... it is after all the end.

It is a typical result. Whenever I go with my "normal" riding partner the AP for a ride is well below what I'd do on my own, and the GP is well above what I could do on my own. I can relate:

http://groups.google.com/group/wattage/ ... c69ab7780c

"Over the 5 months I've had a power meter, however, I've begun to notice that it (the NP algorithm) seems a bit generous."

Re: Rick Murphy's Normanalized Power Thread 8 years, 6 months ago #24459

  • Kirk
  • OFFLINE
  • Administrator
  • Posts: 1102
  • Karma: 2
Interesting stuff. Regarding that wattage group, I haven't really followed it since it switched from topica. My experience there regarding GP associated issues is that there can be an incomplete presentation of data without context, such as how the x^3.9 discussed above isn't presented as having a similar fit quality as a linear fit (if I am wrong on this particular fit issue, someone please correct me)...and more recently I heard about some glycogen stuff which was posted without contrast. So, I'd say that GP/GSS stuff there should be viewed skeptically, plugged into a spreadsheet, verified, and contrast in order to gain perspective on the value/meaning of what is being presented. It can be hard to tell if other relationships have been thoroughly considered. That's my philosophy for any data presentation. Heck, I've seen published data presented to a highly paying audience in a non-entertainment industry setting showing a relationship which looks great in a plot and supported the position of the lecturer...but if I didn't look closely while listening, it would be hard to notice that the R^2 was friggin' horrible (that part wasn't presented)!

Re: Rick Murphy's Normanalized Power Thread 8 years, 6 months ago #24473

  • Nicko
  • OFFLINE
  • Expert Boarder
  • Posts: 86
  • Karma: 0
Kirk wrote:
I... My experience there regarding GP associated issues is that there can be an incomplete presentation of data without context, such as how the x^3.9 discussed above isn't presented as having a similar fit quality as a linear fit (if I am wrong on this particular fit issue, someone please correct me)...

This plot was presented here [url:3amopr5j]groups.google.com/group/wattage/browse_t...4b7ec0fe0bd2d?hl=sv#[/url]
...from 10-20 min bouts of constant-intensity exercise performed by trained male cyclists

[attachment=1:3amopr5j]<!-- ia1 LactateVsVO2.jpg<!-- ia1 [/attachment:3amopr5j]
The suggested power fit seemed incorrect to me, in every single case it failed to recognize the sharp "knee" in the data.
I suggested that lactate should be correlated to "VO2 reserve", i.e 1/(1-VO2/VO2max), and voila, the fit was "perfect", albeit very individual.
[attachment=0:3amopr5j]<!-- ia0 LactateVsVO2reserve.jpg<!-- ia0 [/attachment:3amopr5j]

The problem with the purely power fit, regardless of exponent, is that it is too coarse to fit the "knee" given that it starts from origo. The reciprocal of how close you are to VO2max seems to explain these five individuals lactate levels very good.

My questions and suggestions got the silent treatment :o
'The purpose of computing is insight, not numbers', Hammings motto

Re: Rick Murphy's Normanalized Power Thread 8 years, 6 months ago #24487

  • Kirk
  • OFFLINE
  • Administrator
  • Posts: 1102
  • Karma: 2
It's cool to see some critical thought on that stuff. It's just not clear to me that power fits actually have any real meaning vs. linear fits (let alone using two linear fits and a cut-off), especially when lactate is normally 0.5 to 1.5 at rest anyways.

IIRC, I think there was a glycogen utilization rate vs. relative intensity ^2 plot with a linear fit presented too (one of a few glycogen related plots)? It seems that it actually doesn't matter whether one uses relative intensity ^2 or just plain relative intensity in that relationship...it's all the same eh? Someone please correct me if I got that wrong.

Re: Rick Murphy's Normanalized Power Thread 8 years, 3 months ago #24907

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
I was sent this link:

http://anonymous.coward.free.fr/wattage ... o09-12.csv

and was asked to make one of my pretty plots that appeared earlier in this thread.

Here's the first one that breaks things up into 60 second averages:

[attachment=1:w1v62mr2]<!-- ia1 wiggins_60_seconds.png<!-- ia1 [/attachment:w1v62mr2]

and another that breaks things into 5 minute averages:


[attachment=0:w1v62mr2]<!-- ia0 wiggins_300_seconds.png<!-- ia0 [/attachment:w1v62mr2]

to me, it's pretty clear that this is another example of how, for a given variability, gizmo power doles out bonus points at the lower average powers.

Also, it's pretty clear what is highly correlated with the "GP minus AP" quantity for Bradley Wiggins during his Giro Stage 9 effort, eh?
-kraig

Re: Rick Murphy's Normanalized Power Thread 8 years, 2 months ago #25102

  • djconnel
  • OFFLINE
  • Gold Boarder
  • Posts: 203
  • Karma: 2
I haven't followed all of this but I agree the present approach to calculating rewop dezilamron is flawed in the use of an initial 30-second smoothing. For short-duration efforts, the rewop dezilamron of an interval then becomes highly sensitive to the power which preceded which preceded the interval in the duration of a smoothing time constant, even if that power is well into the aerobic range. For example, consider I do an interval at 200% FTP for 30 seconds. The rewop dezilamron associated with that interval is very strongly affected by whether I was 70% or 0% FTP in the preceding 30 seconds (assuming 30-second smoothing).

Ideally only the anaerobic component of the power should be smoothed. This requires an approximation for the anaerobic power as a function of total power. The idea here is that ATP can be used to spare anaerobic power for short durations, but it is not engaged in fluctuations of power at low intensities. An interval starting at 70% FTP isn't harder than one starting at 10% FTP.

Re: Rick Murphy's Normanalized Power Thread 8 years, 2 months ago #25103

  • JV
  • OFFLINE
  • Platinum Boarder
  • Posts: 1101
  • Karma: 0
djconnel wrote:
An interval starting at 70% FTP isn't harder than one starting at 10% FTP.


You want to make that interesting?

Re: Rick Murphy's Normanalized Power Thread 8 years, 2 months ago #25107

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
JV wrote:
djconnel wrote:
An interval starting at 70% FTP isn't harder than one starting at 10% FTP.


You want to make that interesting?


Oh, I think it would be pretty interesting if GP acolytes acknowledged the fact that at lower average powers and a fixed variability, the GP jobber doles out extra bonus points.

What are your thoughts on this?
-kraig

Re: Rick Murphy's Normanalized Power Thread 8 years, 2 months ago #25111

  • JV
  • OFFLINE
  • Platinum Boarder
  • Posts: 1101
  • Karma: 0
kraig wrote:
JV wrote:
djconnel wrote:
An interval starting at 70% FTP isn't harder than one starting at 10% FTP.


You want to make that interesting?


Oh, I think it would be pretty interesting if GP acolytes acknowledged the fact that at lower average powers and a fixed variability, the GP jobber doles out extra bonus points.

What are your thoughts on this?


I don't have any comments on that. I was interested in Dan's statement, in part because I don't know how it would turn out.

Re: Rick Murphy's Normanalized Power Thread 8 years, 2 months ago #25115

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
djconnel wrote:
<snip>


thanks for your thoughts, Dan.

I think some of what you are noticing fits into some of kirk's comments earlier in the thread.

I'd suggest reading three or four posts down...or whatever you have time for...starting from here:

http://forum.biketechreview.com/viewtop ... 91&#p23791
-kraig

Re: Rick Murphy's Normanalized Power Thread 8 years, 2 months ago #25126

  • djconnel
  • OFFLINE
  • Gold Boarder
  • Posts: 203
  • Karma: 2
I'm not sure I caught the key stuff in that reference. But it's easy enough to come up with a heuristic equation which credibly partitions power into aerobic and anaerobic components for a given power level. Consider power P, aerobic power P?, and anaerobic power P?, where P = P? + P?. I simply need an equation where P? tracks P, then saturates at some CP, where P? takes over. For example:

P? = P / [1 + (P/CP)^n]^(1/n)

for some n. The choice of n determines what fraction of power is aerobic at P = CP. For example, if n = 7, then at P = CP, I get P? = CP / 2^(1/7) = 90.5% CP. That corresponds to a fairly sharp transition from aerobic to anaerobic power, but "feels" about right. Only relatively sharp transitions will reproduce a critical power model result that power ? CP + AWC / duration for durations in the 3-30 minute range.

You can then follow an approach like the following (edited):
1. Smooth power with a time constant relatively small in comparison to 30 seconds (otherwise the power variation through a pedal stroke would matter too much).
2. Partition power into aerobic and anaerobic components.
3. Further smooth the anerobic component with a time constant closer to 30 seconds = P?*.
4. Calculate a special cost function for that anaerobic power (for example proportional to ? P?* exp (P?* / CP) dt).
5. Calculate a cost function for the total power (for example proportional to ? P dt )
6. Add the two cost functions to get effective work.
7. Divide by duration if you want an effective power.

The anaerobic cost function is an effective excess work on top of traditional kJ. The focus here is more on effective work than on effective power. With the rewop dezilamron the focus is on power, which is why krow dezilamron (ie tee-ess-ess) is so munged.

This formula needs of optimization. But it at least doesn't fall for the obvious shortcoming of rewop dezilamron, which is that a completely aerobic effort preceding or following an anaerobic interval can strongly affect the result for that interval due to smoothing. But then it adds an additional modeling parameter (CP) so loses some of the elegance.

One issue with the formula is that constant power no longer results in an effective power equal to the actual power. Restoring this relationship would require a normalization by the appropriate AP-dependent function. But I don't see why this feature would be desirable: the goal is to derive a metric of how hard a ride was. We all agree a ride of any duration done at 110% CP is more than 10% "harder" than one done at CP.

Re: Rick Murphy's Normanalized Power Thread 8 years, 2 months ago #25131

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
djconnel wrote:
I'm not sure I caught the key stuff in that reference.



I wouldn't overthink it too much.
-kraig

New Plot 8 years, 1 month ago #25227

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
After cracking the new SRM format code:

http://forum.biketechreview.com/viewtop ... f=1&t=2453

I have put together this plot that leverages my database:

[attachment=0:19alv10d]<!-- ia0 gizmo_power.png<!-- ia0 [/attachment:19alv10d]

this plot is based on n=750+ rides in my srm database spanning 6 years or so...these files span focused preparation for 20k, 40-ishk TT's, state masters road race championships, 1 hour circut races, 40 minute crits, half-ironmans and full ironmans. The data is what it is. I've sampled the kool-aid early on, and have personally found it wanting...

I've done this same exercise for case after case of PM file databases, and the results seem to be consistent: no new actionable information is gained over actually inspecting the files.

I'm happy to produce the same style of plot (minus promotional material... ;-) ) ) for anyone who has a slew of powertap or srm native format files - for large file databases the number crunching can take a day or so of dedicated CPU time (which is more a reflection of my programming skillz than it is of anything about the complexity of what I database).

If the gizmopower trade-secret guys ever publish their file format, I might even invest some time in producing a script that databases things based on that profit generating format too...you'll have to send me your files and the format spec before I consider that deal, though!!! :-) )
-kraig

Re: New Plot 8 years, 1 month ago #25244

  • Nicko
  • OFFLINE
  • Expert Boarder
  • Posts: 86
  • Karma: 0
kraig wrote:
<snip>
If the gizmopower trade-secret guys ever publish their file format, I might even invest some time in producing a script that databases things based on that profit generating format too...you'll have to send me your files and the format spec before I consider that deal, though!!! )


Kudos to this guy:
http://sourceforge.net/projects/wko2csv/files/

Bulk export to CSV format
'The purpose of computing is insight, not numbers', Hammings motto

Re: New Plot 8 years ago #25323

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
Nicko wrote:
kraig wrote:
<snip>
If the gizmopower trade-secret guys ever publish their file format, I might even invest some time in producing a script that databases things based on that profit generating format too...you'll have to send me your files and the format spec before I consider that deal, though!!! )


Kudos to this guy:
http://sourceforge.net/projects/wko2csv/files/

Bulk export to CSV format


awesome work by that guy!

Now there isn't any excuse for the wko acolytes not to provide their data in csv format - heck, I'd even agree to run that sweet looking csv conversion script on someone's wko files.

FWIW, no one new has taken me up on the offer to run a "slew" of their power files through my databasing script and generate a plot like the one above - y'know, the most recent plot where GP is the dependent variable?

So, I'd encourage folks to send me your "slew" of files to database - in the end it's just numbers, it's not like it's religion or anything.
-kraig

Re: Rick Murphy's Normanalized Power Thread 8 years ago #25374

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
quick update here...

still no volunteers in terms of sharing files for my databasing exercise.

in the HR thread below, I mentioned that I was able to predict my avg HR within 5% or so using average power across all the files in my database - to me, this suggests that physiological strain is somehow related to avg power.

I tried to do the same regression exercise wih the Normanalized power deal, but there was no joy in mudville, so to speak - my own data suggested that there was not a relationship between GP and HR .

What's your avg HR and AP data say?
-kraig

Re: Rick Murphy's Normanalized Power Thread 7 years, 10 months ago #25541

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
Well... a few brave souls have sent me a grip of files to date.

I've managed to get the wko convertor working and also modified my script to allow databasing of the resulting wko to csv converted files.

That convertor script writer guy is awesome!

The most interesting thing to shake out, for me thus far, is the dearth of "value added" information the various "virtual metrics" out there bring to the table.

These "virtual metrics" however, do bring more "confusion" to the table, though -> [url:pft0qdtz]www.biketechreview.com/kdublog/2009/11/training-plot.html[/url]

)
-kraig

Re: Rick Murphy's Normanalized Power Thread 7 years, 10 months ago #25544

  • Kirk
  • OFFLINE
  • Administrator
  • Posts: 1102
  • Karma: 2
Nice. There's something to that sweet plot even if stuff like the sensitivity, specificity, and measurement error is known.

Re: Rick Murphy's Normanalized Power Thread 7 years, 10 months ago #25546

  • Kirk
  • OFFLINE
  • Administrator
  • Posts: 1102
  • Karma: 2
Continuing this theme...in my new environment, in order to properly use a “test” to aid treatment decisions it is essential to consider the specific characteristics of the test in context. Without considering the characteristics of the test, the numbers can lead one down the wrong path...ya know, confuse things.

In terms of the gizmo suite, it seems that there isn't any new information provided. What kind of decision tree do folks use to change training based specifically on the suite of gizmo “test” numbers without knowing and considering the characteristics of the “tests” (noting that one component of error is identified above in this thread)?

Re: Rick Murphy's Normanalized Power Thread 7 years, 10 months ago #25547

  • Kirk
  • OFFLINE
  • Administrator
  • Posts: 1102
  • Karma: 2
Kirk wrote:
Continuing this theme...in my new environment, in order to properly use a “test” to aid treatment decisions it is essential to consider the specific characteristics of the test in context. Without considering the characteristics of the test, the numbers can lead one down the wrong path...ya know, confuse things.

In terms of the gizmo suite, it seems that there isn't any new information provided. What kind of decision tree do folks use to change training based specifically on the suite of gizmo “test” numbers without knowing and considering the characteristics of the “tests” (noting that one component of error is identified above in this thread)?


Maybe this is a better way to approach this question...I'm asking anyone.

When a given “state” or “condition” is being assessed with something from the gizmo suite, what does the decision making process look like for pre-test probabilities ranging from high to low (ie. the likelihood of a condition being present or absent before considering gizmo suite numbers) in the absence of sensitivity and specificity characteristics from which positive and negative predictive values can be considered?

Re: Rick Murphy's Normanalized Power Thread 7 years, 10 months ago #25548

  • Kiwi Coach
  • OFFLINE
  • Junior Boarder
  • Posts: 23
  • Karma: 0
I don't follow the numbers too closely as there is too much of a garbage in garbage out risk going on. I do look at trends as I do believe that if one is increasing fitness without letting form drop too low then it's something extra to add to the mix of increasing power, resting HR, hours of sleep, behaviour etc to ensure training is headed in the right direction.

Re: Rick Murphy's Normanalized Power Thread 7 years, 10 months ago #25553

  • liversedge
  • OFFLINE
  • Fresh Boarder
  • Posts: 1
  • Karma: 0
kraig wrote:

That convertor script writer guy is awesome!

Aw shucks. I'd say I was more OCD than awesome. It was a labour of pure unadulterated obsession. BTW, I have lots of WKO files, lots. PM me and I'll work out the best way to get to you.

thisisindexed.com is now my new favourite blog, thanks for the tip!

Re: Rick Murphy's Normanalized Power Thread 7 years, 10 months ago #25586

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
kraig wrote:
Well... a few brave souls have sent me a grip of files to date.


Quick update here: I'm now up to around 2000 files across individuals collected and databased.

These data continue to tell the story that:

1) raise the left
2) fill the right
3) scatter the plot

are nice, independent variables.
-kraig

Re: Rick Murphy's Normanalized Power Thread 7 years, 8 months ago #25888

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
kraig wrote:
kraig wrote:
Well... a few brave souls have sent me a grip of files to date.


Quick update here: I'm now up to around 2000 files across individuals collected and databased.

These data continue to tell the story that:

1) raise the left
2) fill the right
3) scatter the plot

are nice, independent variables.


...for kicks, I used the aforementioned independent variables across all files in my database to predict one of the gizmo metrics...and, as I have experienced over the years at intermittent times of curiosity/"etc...", I found that this particular gizmo metric (once again) doesn't bring any new information to the table...r^2 was greater than 0.9...so, yeah, "raise, fill, scatter".
-kraig

Re: Rick Murphy's Normanalized Power Thread 7 years, 7 months ago #25969

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
___
[attachment=0:13nafms8]<!-- ia0 fun_with_plots.png<!-- ia0 [/attachment:13nafms8]
___
-kraig

Time to create page: 0.98 seconds

Poll

Which type of tire is more aerodynamic?