Homebrewed "TSTWKT"

[acoggan]

The screenshot is from Excel. We are not part of the inner circle so we didn't have anything more to go on than any other member of the public.

Congrats on "cracking the code". Given how many hints have been dropped, I'm surprised that more people haven't attempted to do so.

As somebody who is not a member of the inner circle, I'd say that you qualify as an impartial observer...I'd therefore be interested in knowing whether you have found this approach useful or not, and in what way(s).

P.S. It looks to as if your modeling might be improved by including some "dummy" values for the initial CTL/ATL/TSB, i.e., as of 1/1/2006. Either that, or you need to include data from, say, the last quarter of 2005 in the calculations.

 

[acoggan]

I must admit, all this "secret sauce" nonsense drives me a bit bonkers, and seems counter to the spirit I'd expect from this sort of enterprise. If TSS, IF, VI, etc., were all opaque and unexplained, I'd treat them with hugely more suspicion. If the formula is "black magic," how am I supposed to critically evaluate the model in question? How can I understand the assumptions that went into it, and its possible limitations?

You can blame Paul Smeulders for the fact that development of this idea has been conducted in private (I won't say secret, since to me at least that implies that I/we are somehow obligated to share such ideas with the general public, which is obviously not true). That said, I agree with you that complete transparency is ideal, and rest assured that when this thing does go public a full explanation of origins of the idea, how the calculations are performed, etc., will be made available. In fact, as soon as I send this I'm off to start writing an article entitled "The science behind the Performance Manager". (Unfortunately, Training Manager is what Suunto calls the software that comes with their T6 running watch/computer, hence the name change.)

 

[WarrenG]

Originally Posted by WarrenG
Your (RDO's) approach is better since not all levels of intensity need the same amount of work, nor recovery time.

Hence, the use of NP and TSS. That's the whole point.

Ah, so you think a ride or group of rides with a given NP and TSS is just like any other ride or group of rides with similar NP and TSS? Good luck with that approach.

 

[cclarke]

You can blame Paul Smeulders for the fact that development of this idea has been conducted in private (I won't say secret, since to me at least that implies that I/we are somehow obligated to share such ideas with the general public, which is obviously not true).

Was it this exchange? From Topica Wattage:

"Yeah, I agree with this very much...with all the problems and timing
issues around the IPR rights, probably the most effective $trategy
really is to be first to market with extreeemely high-value offerings,
and keeping a few things secret enough.

..and make the rest of them, therefore, always look like "followers".
Keeping it secret also means you won't get others even ASKING for
licenses prior to the first to market release of your preferred
licensee, either...so that the others are forced to always be behind,
and you aren't seen to be unreasonably exclusive...AND it may even
result in increasing the VALUE of the license to the competitors when
they KNOW the other guy (preferred licesnee) is getting too far
ahead....

when things are too secret or undisclosed, I think people doubt its
value or validity, but that's the line inventors and
people-hoping-to-be-even-meagerly-rewarded have to walk....and they have
to get good sales people or advocates. You have these, in spades, beyond
yourself, incidentally.

I'm struggling myself on what's the best way to feed myself over the
winter from my GPS application, I can't work this much for nothing. It's
looking fine though......waiting for that horrible, inevitable day where
nothing goes right and it looks like it will sink...so far, it floats.

Paul.

NOW everyone is going to be ****** at me if this causes you to go too
far underground with your ideas..."


Speaking as someone way, way, way outside the loop (to adopt Paul's manner of expression) and as a pure hobbyist, the process of thinking about the ideas behind the performance manager has led me to many avenues of inquiry that I probably never would have pursued if the product just appeared in final form. Because I get a lot of satisfaction out of learning new things outside my area of expertise, I am thankful for the opportunity. Even the math part.

And, of course, I'm looking forward to its eventual public release.

-Chris

 

[whoawhoa]

Ah, so you think a ride or group of rides with a given NP and TSS is just like any other ride or group of rides with similar NP and TSS? Good luck with that approach.

Adaptation wise, no. Stress wise, yes. That's why TSTWKT is a rough metric for "form"-the output depends on the input. Lots of tss points garnered through high volume, l2/3 training may result in a different type of "form" than lots of l5/6, but the system's job is only to predict when a peak will occur.

 

[cclarke]

Hi Warren-

I don't think anyone believes you should ignore the specifics of the workouts that comprise the training load. I think those details will always matter.

-Chris

 

[acoggan]

so you think a ride or group of rides with a given NP and TSS is just like any other ride or group of rides with similar NP and TSS?.

I doubt that Tom does, as he's too smart to make that simple mistake.

Good luck with that approach.

Some food for thought:

1) the most common metric for quantifying training load is TRIMP, which not only makes no distinction between training at different intensities, it gives you absolutely zero credit for any training performed at >~100% of VO2max (since TRIMP is based on percentage of heart rate reserve, which is "capped" at 100%). Yet, using TRIMP as the input function in an impulse-response model allows you to predict day-to-day variations in performance with 60-90% accuracy.

2) the AIS tried a more complicated, less "global" approach to quantifying training load than TRIMP (or TSS), but gave up because it proved unworkable, i.e., they were unable to ascertain any relationship between training and performance.

 

[acoggan]

Was it this exchange? From Topica Wattage:

"Yeah, I agree with this very much...with all the problems and timing
issues around the IPR rights, probably the most effective $trategy
really is to be first to market with extreeemely high-value offerings,
and keeping a few things secret enough.

..and make the rest of them, therefore, always look like "followers".
Keeping it secret also means you won't get others even ASKING for
licenses prior to the first to market release of your preferred
licensee, either...so that the others are forced to always be behind,
and you aren't seen to be unreasonably exclusive...AND it may even
result in increasing the VALUE of the license to the competitors when
they KNOW the other guy (preferred licesnee) is getting too far
ahead....

when things are too secret or undisclosed, I think people doubt its
value or validity, but that's the line inventors and
people-hoping-to-be-even-meagerly-rewarded have to walk....and they have
to get good sales people or advocates. You have these, in spades, beyond
yourself, incidentally.

I'm struggling myself on what's the best way to feed myself over the
winter from my GPS application, I can't work this much for nothing. It's
looking fine though......waiting for that horrible, inevitable day where
nothing goes right and it looks like it will sink...so far, it floats.

Paul.

No. In fact, the thinking he expressed in that post is about as foreign to my nature as you can possibly imagine. Rather, the reason that the development of the Training Manager has taken place out of the public view is simply because Paul attempted to make $$$ off of ideas that I had been giving away for free, despite knowing that I considered that unethical. So, as the saying goes: once burned, twice shy.

 

[cclarke]

No. In fact, the thinking he expressed in that post is about as foreign to my nature as you can possibly imagine. Rather, the reason that the development of the Training Manager has taken place out of the public view is simply because Paul attempted to make $$$ off of ideas that I had been giving away for free, despite knowing that I considered that unethical. So, as the saying goes: once burned, twice shy.

Too bad. That's what I originally thought happened, although I couldn't remember who had coopted your ideas, but came across his comments and hoped for the best.

-Chris

 

[RapDaddyo]

Adaptation wise, no. Stress wise, yes. That's why TSTWKT is a rough metric for "form"-the output depends on the input. Lots of tss points garnered through high volume, l2/3 training may result in a different type of "form" than lots of l5/6, but the system's job is only to predict when a peak will occur.

From what I know about TSTWKT, I agree with this statement. I think what I am fooling around with is more along the line of delving into the adaptation piece. Not as an alternative to TSTWKT, but rather as supplemental information. I'm pretty close to being able to plan rides to realize multiple goals, including NP, IF, TSS and ride time by level. So, now the question is what metrics to use for ride planning. I want to incorporate adaptation goals as well as total training stress goals. Of course, the Cray computer that I have to drag around does sort of hurt my w/kg (not to mention the power cord that keeps getting tangled in traffic). I need to move to SFO so I can ride the trolley routes and hook up to their overhead power cables.

 

[Tom Anhalt]

Congrats on "cracking the code". Given how many hints have been dropped, I'm surprised that more people haven't attempted to do so.

As somebody who is not a member of the inner circle, I'd say that you qualify as an impartial observer...I'd therefore be interested in knowing whether you have found this approach useful or not, and in what way(s).

P.S. It looks to as if your modeling might be improved by including some "dummy" values for the initial CTL/ATL/TSB, i.e., as of 1/1/2006. Either that, or you need to include data from, say, the last quarter of 2005 in the calculations.

How sure are you that Squint cracked the code? It seems to me that I was able to very easily enter the proper "code" into the Excel formula bar....

I know that I personally can't tell if it's the correct method just by looking at a screenshot...

Just curious (and just trying to maintain my exclusivity as the "code cracker )

 

[Tom Anhalt]

I doubt that Tom does, as he's too smart to make that simple mistake.

Thanks for the compliment ...but I really am just a "caveman" with most of this stuff.

I couldn't have put it any better than what whoawhoa said:

Adaptation wise, no. Stress wise, yes. That's why TSTWKT is a rough metric for "form"-the output depends on the input. Lots of tss points garnered through high volume, l2/3 training may result in a different type of "form" than lots of l5/6, but the system's job is only to predict when a peak will occur.

It's a VERY subtle, yet important distinction about what something like the Performance Manager can or can't do.

 

[acoggan]

How sure are you that Squint cracked the code? It seems to me that I was able to very easily enter the proper "code" into the Excel formula bar....

I know that I personally can't tell if it's the correct method just by looking at a screenshot...

Just curious (and just trying to maintain my exclusivity as the "code cracker )

Well, you're right, I am just guessing based on the screenshot. However, I don't take his comment about not being able to enter the relevant formulae directly into Excel as meaning he did or did not "crack the code". After all, it was only after Rick Murphy provided me w/ an alternative expression that could be used that I started working on this idea again...

 

[Squint]

First of all, I have to give most of the credit to my friend who did most of the reading and coding.

I'm not that good with Excel so when he told me the formula couldn't be entered into the formula bar, I didn't have any suggestions other than to shift to Plan B which was to apply the formula in the form of Basic code instead. I gave him a rough outline of how the code should be written and he wrote it in OpenOffice which I then translated into VBA for Excel. All this took place within the span of 48 hrs. I was never that motivated alone but my friend couldn't stand it and immediately began working on it.

I haven't had the time to thoroughly check his work but the past few months that we've used it, it seems to work. When I've had good form this season and bad form, it's been reflected in TSB. We've used it to give us an idea when to rest or keep building or start the next training block. I've also used it to enter hypothetical training data to see if CTL continues to climb.

I did do a block of mostly L2/3 high-TSS workouts and performance, subjectively and as measured by 5 & 20 min tests, declined. I've replaced the L3 workouts with 2x20s at no less than 95% of FTP and 1x60 in L4 and performance seems to be on the rise though I haven't done any testing yet.

My L4 workouts are almost always lower TSS (~100) compared to tempo rides (~165 TSS) so it seems I have to keep that in mind when looking at TSB. Because all TSS points are not equal, TSB may not always be equal. At least that's the impression I'm getting.

BTW, we call it Rage Manager.

 

[acoggan]

First of all, I have to give most of the credit to my friend who did most of the reading and coding.

I'm not that good with Excel so when he told me the formula couldn't be entered into the formula bar, I didn't have any suggestions other than to shift to Plan B which was to apply the formula in the form of Basic code instead. I gave him a rough outline of how the code should be written and he wrote it in OpenOffice which I then translated into VBA for Excel. All this took place within the span of 48 hrs. I was never that motivated alone but my friend couldn't stand it and immediately began working on it.

I haven't had the time to thoroughly check his work but the past few months that we've used it, it seems to work. When I've had good form this season and bad form, it's been reflected in TSB. We've used it to give us an idea when to rest or keep building or start the next training block. I've also used it to enter hypothetical training data to see if CTL continues to climb.

I did do a block of mostly L2/3 high-TSS workouts and performance, subjectively and as measured by 5 & 20 min tests, declined. I've replaced the L3 workouts with 2x20s at no less than 95% of FTP and 1x60 in L4 and performance seems to be on the rise though I haven't done any testing yet.

My L4 workouts are almost always lower TSS (~100) compared to tempo rides (~165 TSS) so it seems I have to keep that in mind when looking at TSB. Because all TSS points are not equal, TSB may not always be equal. At least that's the impression I'm getting.

BTW, we call it Rage Manager.

If you want, I'd be happy to double-check your calculations to see if we're really on the same page...just drop me an email at acoggan at earthlink dot net.

Rage Manager...nice! That makes me think of what Dave Harris had to say about it just the other day:

"If my CTL is 135 and TSB 35 the cranks are in danger regardless of composition of training!"

 

[WarrenG]

Some food for thought:

...Yet, using TRIMP as the input function in an impulse-response model allows you to predict day-to-day variations in performance with 60-90% accuracy..

60-90% accuracy? Hmmm. I think there are ways to do it much better than that.

2) the AIS tried a more complicated, less "global" approach to quantifying training load than TRIMP (or TSS), but gave up because it proved unworkable, i.e., they were unable to ascertain any relationship between training and performance.

IOW, trying to "quantify" the training load was/is unworkable to a level of accuracy that would have been useful for the AIS?

 

[WarrenG]

Originally Posted by WarrenG

Ah, so you think a ride or group of rides with a given NP and TSS is just like any other ride or group of rides with similar NP and TSS? Good luck with that approach.

Adaptation wise, no. Stress wise, yes. .

You are mistaken about the stress(es). There are various types of stress, and they do not all have the same impact on a given performance nor the type of training you could do next (during that session, that block of days, that week...). And it will vary between individuals.

 

[acoggan]

60-90% accuracy? Hmmm. I think there are ways to do it much better than that.

You misunderstand. The modeling approach can account for 60-90% of the day-to-day variation in performance, which in itself is usually <5% (when underlying fitness is itself constant). IOW, via modeling you can predict how someone will perform on any given day to w/in 0.5-2% (i.e., 10-40% of 5%). Seems like, oh, Leipheimer might have benefitted from such knowledge when he was trying to set himself up to defend his title in Germany...if he had, he might not have lost 42 s already.

IOW, trying to "quantify" the training load was/is unworkable to a level of accuracy that would have been useful for the AIS?

As Dave Martin described it to me, they were trying to model the relationship between training and performance based on the notion of "spheres of influence", e.g., the idea that training at intensity/duration X would improve your performance mostly at intensity/duration X, but also, to a lesser extent, and intensities/durations Y and Z. When training was broken down this way/to this degree, though, the modeling approach fell apart, i.e., it wasn't possible to show any relationship between training and performance (which, as discussed above, can be done using other measures of training load).

 

[peterpen]

... Seems like, oh, Leipheimer might have benefitted from such knowledge when he was trying to set himself up to defend his title in Germany...if he had, he might not have lost 42 s already. ...

Leipheimer lost 42s because he's a tiny dude and it was super windy + dumping rain when he went (unlike most of the competition.)
Plus, he's a bit of a sissy in the wet.

But yeah, I bet he'd love to have something to quantitatively manage his form, which seems to be all over the place this season.

 

[Tom Anhalt]

First of all, I have to give most of the credit to my friend who did most of the reading and coding.

I'm not that good with Excel so when he told me the formula couldn't be entered into the formula bar, I didn't have any suggestions other than to shift to Plan B which was to apply the formula in the form of Basic code instead. I gave him a rough outline of how the code should be written and he wrote it in OpenOffice which I then translated into VBA for Excel. All this took place within the span of 48 hrs. I was never that motivated alone but my friend couldn't stand it and immediately began working on it.

I haven't had the time to thoroughly check his work but the past few months that we've used it, it seems to work. When I've had good form this season and bad form, it's been reflected in TSB. We've used it to give us an idea when to rest or keep building or start the next training block. I've also used it to enter hypothetical training data to see if CTL continues to climb.

I did do a block of mostly L2/3 high-TSS workouts and performance, subjectively and as measured by 5 & 20 min tests, declined. I've replaced the L3 workouts with 2x20s at no less than 95% of FTP and 1x60 in L4 and performance seems to be on the rise though I haven't done any testing yet.

My L4 workouts are almost always lower TSS (~100) compared to tempo rides (~165 TSS) so it seems I have to keep that in mind when looking at TSB. Because all TSS points are not equal, TSB may not always be equal. At least that's the impression I'm getting.

BTW, we call it Rage Manager.

Hmmm...I'm going to guess that you guys coded up one of Busso's equations, right?

IIRC, Kraig W. tried that with less than stellar results, right Andy?

When I looked at the Busso work, one of the things that seemed to be difficult in applying it was the necessity to derive certain constants from the data itself...

My "Aha!" moment was when I realized that, based on some of his reasonings for the basis of NP, if Andy was going to apply some sort of impulse-response model it would be a lot simpler than the Busso or Bannister stuff.

I like "Rage Manager" too ...one thing I've realized about the name "Performance Manager" is that you can't just shorten it to PM since that will get confused with PM for Power Meter.