Re: Skiba Formula for AWC calculation

[Mark Liversedge]

Sounds good -- fancy raising a featurer request -- we can slot this into 3.1 if Dr Phil is cool with it.

Thanks,

Mark

On Monday, June 3, 2013 6:48:30 PM UTC+1, Armando wrote:

Dr. Phil Skiba published a research paper to calculate W' expenditure and reconstitution:

http://www.ncbi.nlm.nih.gov/pubmed/22382171

Do you think we should look into getting his permission to use this to calculate a more accurate AWC value in GC?

[Dr. Philip Skiba]

Hey Mark,

Happy to have you guys include it. The paper has made a pretty big splash in the elite cycling world and I think including it would help you get more people onboard with GC.

Best,

Phil

[Mark Liversedge]

You are a hero Dr Skiba, we stand on your shoulders.

[Mark Liversedge]

Dr Skiba has kindly shared some implementation details and so I've added the possibility to plot the W' balance on the ride plot. The red line is W' balance, which always starts from 0. I will look to allow users to specify their W' alongside their CP in preferences as a future update.

I'm interested to know what key metrics we think would be useful?

I reckon plotting 'how deep you went' and totals for 'utilisation' and 'replenishment' would be good.

But any others?

Mark

PS: As ever -- we all owe a huge debt of gratitude to Dr Phil. Thanks!

[Mike Patton]

Great stuff!

I would propose having a metric that would get at how many "matches" you've burned as a way of looking at 'how deep you went'. An individual "match" could be defined as a distinct period of W' depletion. The issue is that you'd have to smooth something to avoid classifying periods of time where power goes briefly below CP as being a bunch of little matches, when in fact they amount to a slow-burning match that is very burnt. Perhaps the absolute value of W' could be used to avoid that issue. For example, I'd count about 11 matches in the graph you've posted just visually, but I'm not sure how exactly you would code that.

Mike

[Dr. Philip Skiba]

Hi Guys,

Tried to post this a few minutes ago but got an error message!

Thanks for the kind words. I'm always happy to support Da Cheetah.

First things first: the academic credit for my work needs to be equally shared among my co-authors (both present and future, as there are more papers coming). They are among the best physiologists in the world, and we work as a team. I'd ask that if you label it with my name, please put "Skiba et al 2012".

Ok, on to the implementation. I'm a couple of revisions behind I think, so please excuse me if anything I mention below has already been addressed.

1) In terms of deciding on a CP and W'. I would suggest at least 3 tests. One at 2-3 min, one at 5-7 min, one at 12-20 min or thereabouts. (RaceDay Apollo will allow you to use as many as 2 (not recommended) up to as many as 5.) I'd probably avoid going under 2 min, however, I have made some very good models from field data down as low as 1.5 minutes in some circumstances.

2) You can also try auto-extracting CP and W' from your mean max power plot. You can get SOMETIMES get *serviceable* results if you ride REALLY hard with some frequency. However, you need to be really seriously riding across a wide variety of power outputs, and often. Otherwise it is way off, and leads to poor data quality and crummy models . In my experience, most people don't do this, particularly triathletes :-). The other important challenge is the time windows...you don't want to pull a 5 min best power from 2 years ago if you haven't been training much lately, for obvious reasons.

3) I'd suggest taking out AWC and making it W'. Or maybe " W' (AWC) ". That's because AWC is old language at this point, and there is enough data out there now indicating that the W' is not strictly anaerobic in nature. (If you are interested, I recently co-authored a paper with Dionne Noordhof and Jos De Koning on the CP model and some other models that try to quantify "anaerobic" work. It's very applied and you might like it.)

4) We're calling the model the W'BAL model, for W' balance. The BAL is subscripted.

I've some more papers in review at the moment, and I'll pass along the details as I am able, as they suggest interesting interesting things to add to Cheetah. However, journals get very shirty if you put stuff on the internet before they have a chance to publish. It is really important to me that my work continue to be reviewed, critiqued and improved upon by my fellow scientists before it is publicly released. So, I just need the stuff to be in print first to ensure the best quality science.

I also see Mike's comment below, who I had the pleasure of meeting at ACSM last year. Hey Mike!

Best Wishes to All,

Phil

[Mark Liversedge]

On Thursday, October 31, 2013 5:51:49 PM UTC, Mike Patton wrote:

I would propose having a metric that would get at how many "matches" you've burned as a way of looking at 'how deep you went'. An individual "match" could be defined as a distinct period of W' depletion. The issue is that you'd have to smooth something to avoid classifying periods of time where power goes briefly below CP as being a bunch of little matches, when in fact they amount to a slow-burning match that is very burnt. Perhaps the absolute value of W' could be used to avoid that issue. For example, I'd count about 11 matches in the graph you've posted just visually, but I'm not sure how exactly you would code that.

Cool idea. I have no idea how we'll do it, but thats the kind of challenge we like.

It can't be that hard, right?

I'll add it to the feature request. 

[Mike Patton]

Cheers, Mark.

Let me know if I can help at all. I'm not much of a programmer but am reasonably familiar with developing algorithms.

Hi Dr. Skiba. The pleasure was mine! Looking forward to all this new modeling you're hinting at.

Mike

[Mark Liversedge]

I've just pushed a couple of fixes. I was using a constant for tau instead of calculating it (!). And now show it on the chart. I've found it useful to have two ride charts, one for general analysis and one to show power vs W' bal.

[Mark Liversedge]

It struck me in conversation with friends that the significance of  W' balance may be lost on a few folks.

It may appear to be some arcane science mumbo jumbo that doesn't really mean anything to the average joe.

Fortunately this video from Dr Phil helps to explain it and why its so useful.

http://www.youtube.com/watch?v=86Sw3vOCq9U

Regards,

Mark

[AndyB]

Looks very interesting.

Please could you explain the Y axis? I seem to start at 0 and get down to -20,000 -- I thought it would head towards 0 as I get closer to exhaustion.

I guess this stuff isn't quite finished. If you'd like any help testing I'd be happy to get involved. So far I've found a very fatiguing cafe stop...

[Mark Liversedge]

I hadn't added the ability for the user to set their W'.

It is in the latest repo, got pushed yesterday.

So if a user maintains their W' with their CP in preferences it will start from W' rather than 0.

Here is an example:

[AndyB]

Thanks. It was working fine, but I'd set W' in Kj (from the CP chart) rather than j. Changing that sorted it.

Are you interested in the ride with the strange cafe stop (image attached)?

[Mark Liversedge]

You have a gap in recording. See Ride Editor -> Anomalies.

[Martin Wifling]

Is it possible already in GC3.0 to calculate W´or AWC? I have calculated CP, but AWC? Or is it to be done manually, as have not found a predefined metric for AWC/W´?
Or do I miss something?

However, if you can get it working this would be a killer functionality. I always struggle on how many interval repeats should be effective, this could give me a good target estimation.

[Mark Liversedge]

There is an estimate on the CP chart based upon the bests shown (if power is being charted).

It adjusts as you select different date ranges.

I cannot vouch for its accuracy and also the testing protocol needs to be carefully executed.

I believe Dr Skiba is going to publish something on this reasonably soon -- and we will implement something to match it when he does.

Cheers,

Mark

PS: I eyeballed old and current data to see how deep I went on hard / exhausting rides to validate the values I got from the CP chart and they were reasonable.

[Martin Wifling]

Thank you Mark, I have been blind ...

Cheers, Martin

[Mark Liversedge]

OK.

I need a definition of a match in terms of W' depletion and power above CP !

Cheers,

Mark

[Mark Liversedge]

If it helps to see examples, here are 4 rides with the peaks and troughs in the curve marked as blue squares. I haven't done any smoothing etc as thats really what this question is about.

Ride 1

Ride 2

Ride 3

Ride 4

[Mark Liversedge]

I've now updated the find intervals dialog to find matches using a user defined minimum kJ cost (in terms of W'bal).

Here are a couple of screenshots to explain (from separate rides!):

Cheers,

Mark

[Pete from AUS]

Are there potential refinements from considering W'BAL for metrics such as xPower and BikeScore?

Eg power relative to W'BAL might indicate a higher effort when W'BAL is low compared to a greater power output when more W'BAL is available?

[Mark Liversedge]

On Thursday, 3 April 2014 04:09:33 UTC+1, Pete from AUS wrote:

Are there potential refinements from considering W'BAL for metrics such as xPower and BikeScore?

Eg power relative to W'BAL might indicate a higher effort when W'BAL is low compared to a greater power output when more W'BAL is available?

OH YES.

The W'bal can be considered a 'system' measure, and as such would reflect overall fatigue -- so efforts at lower W'bal should, in theory, be more stressful than the same efforts at a higher W'bal.

Watch this space ...

Mark 

[pbi...@gmail.com]

Hi Coach Phil,

I'm currently working on some related research to your W'Bal and was wondering if you could help me understand it a bit better.

Best,

Phoebe

[Rob Stanley]

Hi Phoebe, 

Did you ever receive a response from this? 

I would be interested to chat to you about your questions here. 

Thanks, 

Rob 

[Gregory Mocko]

Hey Rob,

Phoebe was a graduate student under my advisement. She has taken a job and is no longer actively studying this area. Our research group continues to work in this area however. So I will try to answer for her. 

No we have not received a response on this forum. We did contact Phil Skiba outside of the forum and we were told that due to contractual obligations he could not provide additional information on the model. We believe this is because he is / was working with a company to further develop and implement / refine the model. 

As a note, he contributed to the Breaking2 Nike project and we recently found a US patent application:  US20170266498A1. This application seems to implement a model (similar to Skiba 1 and Skiba 2) in wearable technology.

We recently presented our work at ISEA 2018 and are continuing testing to model both expenditure and recovery.  The best way to contact is Greg Mocko (gmo...@clemson.edu) and Vijay Sreedhara (vsr...@g.clemson.edu).

All the best,

Greg