Franz, Renato, Duncan
Thanks so much for your contributions. Excellent thread for a lot of us to learn from.
Cheers
Franz, Renato, Duncan
Thanks so much for your contributions. Excellent thread for a lot of us to learn from.
Cheers
Dear all,
We would like to thank everybody very much for the comments on our paper and on related topics - this was quite enlightening for us.
Below is a short summary of the discussion, for the convenience of anybody who might find this thread, formulated as Q&A.
Please comment if you feel your contribution misrepresented.
Also, once again we would like to point out that our paper is still unpublished and has yet to pass peer review, so until then please do not make personal decisions based on our results.
Apart from that, talk to you later.
Bye
Franz and Duncan
There was lots of discussion on whether certain athletes X will be able to achieve performances Y independently of our paper, which was quite interesting but not the main topic of our paper, so we will not comment much on that (it's still in the thread).
For those to whom this claim seems strange: our paper is about predicting performances for a large range of athletes, but not about a few athletes X specifically. This is why we do not discuss say the personal history of say Kenenisa Bekele or how he was trained in detail. Whenever we make a prediction about a specific athlete X this has to be read as "if athlete X is, as far as the number tell us, similar to most other athletes we looked at, then ..."
Regarding the scientific approach, here are a couple of objections we encountered, which we believe to be invalid:
Objection: Mathematics and statistics are inappropriate to model physiology.
Response: We think application is difficult, but not by itself problematic. One just needs to be very careful about what one concludes, and avoid to claim formulae to be correct without proper reason. Sometimes people aren't and don't, which may contribute to the distrust.
(this is how the discussion with Renato started, page 3)
O: Prediction X cannot be, thus the paper is incorrect. Or, Prediction X is right, thus the paper is correct.
R: Neither of these is follows logically - a statistical/inductive theory is correct by being correct most of the time. So being correct one time or being wrong another time does not mean the whole theory is right or wrong - we only claim to be accurate most of the time (and our paper has the exact numbers on this).
O: A model cannot be right if it does not include all variables.
R: It cannot be perfect, but it can be accurate (see page 7). Just as you can roughly estimate when your train will reach the station without knowing the names of all passengers.
Regarding the methodology of our analysis:
O: World elite athletes are special, so the model does not apply to them.
R: In terms of everything we see, they are indeed very special (see page 5). Though so far we have no clear evidence whether our model does, or does not apply to them. To see this, future research would be necessary.
Renato also points out that correlation between two close events drops as athletes become more elite. We checked and we seem indeed to be able to confirm this quantitatively.
O: Predictions are made from athletes on the top 25 percentiles, this is problematic when they are made for athletes below.
R: In our paper we checked that this makes the predictions only better, or at least not worse.
O: Ventolin's calculator is the best.
R: On average, it seems not to be better than a random guess and is worse than Riegel's formula.
It is 3 times worse in terms of RMSE than the method proposed in the paper when evaluated on the top 25% of athletes.
Moreover, the method does not improve greatly when restricted to events between 800m and 10km.
(this is not in the original paper and we provided no code, but you will be able to verify it after code release when the paper is published)
O: There are some events which are more difficult, or impossible to predict with the current methodology; for example Renato claims the marathon would be problematic.
R: Our paper gives prediction errors by event, though the original version did not include the "guessing" baseline so it did not allow a final conclusion on this.
We checked and classical methods such as Purdy allow a fairly good prediction for all events, including the Marathon (which seems fine), better than a random guess - except, strangely, 400m.
Our new method is better than the old ones and also allows for a better-than-guessing prediction on 400m, though this is still the distance where prediction seems hardest.
O: It is problematic to put performances on road races and track races together.
R: Indeed, this could be problematic in principle. However, see Renato's posts for reasons why it should not cause issues (page 5). We checked quantitatively, and Renato was right, there is nothing that breaks. We'll include this in the final version.
an interesting thread, the most interesting nugget gleaned being that Renato thinks BK was closer to a potential 26:00 flat!?!!! I am doubtful, but of course Renato has more credibility to make the claim.
I have nothing to add, other than to support the view that the marathon, due to event duration, has far more variables which therefore makes it harder to achieve maximum theoretical potential performance whenever attempted. The main culprits being fuel consumption and dehydration.
I'll also add that the empirical evidence (i.e. actual race results) suggests we're decades away from a 2:01, or even a sub 2:02. So really, I think you need to find out exactly what those missing variables are to account for the discrepancy in marathon times.
Finally, you've made a rather silly, illogical argument above- that is "you can roughly estimate when your train will reach the station without knowing the names of all passengers". As pertains to this discussion, I would revise that to:
"you can roughly estimate when your train will reach the station without knowing the numbers/weight of all passengers".
But really, back to the drawing board. Your challenge is (or should be) to identify and quantify the missing variables. We're nowhere near 2:00:36 and everyone on here knows it.
Duncan Blythe and Franz Kiraly wrote:
O: Ventolin's calculator is the best.
R: On average, it seems not to be better than a random guess and is worse than Riegel's formula.
It is 3 times worse in terms of RMSE than the method proposed in the paper when evaluated on the top 25% of athletes.
Moreover, the method does not improve greatly when restricted to events between 800m and 10km.
(this is not in the original paper and we provided no code, but you will be able to verify it after code release when the paper is published)
Not surprising.
Duncan Blythe and Franz Kiraly wrote:
O: Ventolin's calculator is the best.
R: On average, it seems not to be better than a random guess and is worse than Riegel's formula.
It is 3 times worse in terms of RMSE than the method proposed in the paper when evaluated on the top 25% of athletes.
Moreover, the method does not improve greatly when restricted to events between 800m and 10km.
(this is not in the original paper and we provided no code, but you will be able to verify it after code release when the paper is published)
agreed
but i would like ventolin to respond to this point
he should have right of reply
renato truely a joke .
why give him the time of day....
Cera.............cera and more...
see what the best epo could do
for bekele ,
gebreselasie revival at older age.
wariner,
alan webb ,
at time when epo use had to be
limited could go all out with this
stuff until summer 2008.
when all of them never quite same.
cera best epo for raising and maintaining hematocrit with ease.
dont think bekele would of had
burned fat efficiently enough to get under 2 hours
This is incorrect. There is no evidence that we are decades away from a sub 2:02. The record is currently 2:02:57 and in the past 10 years, 2.5 minutes have been taken off the record. At most, your evidence suggests that it would take ONE decade.
Why are you doubtful that Bekele was close to 26:00? If you saw his record, you would know that he could have run much faster.
Duncan Blythe and Franz Kiraly wrote:O: Ventolin's calculator is the best.
R: On average, it seems not to be better than a random guess and is worse than Riegel's formula.
It is 3 times worse in terms of RMSE than the method proposed in the paper when evaluated on the top 25% of athletes.
Moreover, the method does not improve greatly when restricted to events between 800m and 10km.
(this is not in the original paper and we provided no code, but you will be able to verify it after code release when the paper is published)
i tried to be kind, but your analysis turned out to be drivel
your paper & conclusions were total drivel
people want real world answers
mo ran 50.8 finish in a 14+k run prior to monaco '13 1500
i said calculator with some insight, reckoned that indicated he was
~ 3'28
what is your nonsense going to predict for 50.8 in a 14+ with ole 12'53 & 26'46 pbs ???
impress me !!!
i even gave you best mathematical tables you will likely ever calculate for track sprint/distance :
https://2008olympictrialsakatommyleonard.shutterfly.com/filecabinetsee
"Ventolin worksheet"
ventolin^3 wrote:
....
mo ran 50.8 finish in a 14+k run prior to monaco '13 1500
i said calculator with some insight, reckoned that indicated he was
~ 3'28
what is your nonsense going to predict for 50.8 in a 14+ with ole 12'53 & 26'46 pbs ??? ....
Vent .. so you included those pb's in your calculations?
What Ventolin is saying in essence:
"I tried to be kind, but you had the unwarranted audacity to disagree with me. Therefore, your analysis is drivel and I will be kind no longer."
wtfunny wrote:
ventolin^3 wrote:
....
mo ran 50.8 finish in a 14+k run prior to monaco '13 1500
i said calculator with some insight, reckoned that indicated he was
~ 3'28
what is your nonsense going to predict for 50.8 in a 14+ with ole 12'53 & 26'46 pbs ??? ....
Vent .. so you included those pb's in your calculations?
what part of
with some insight
are you unable to comprehend ???
Following the recent discussion with Canova and co.
we have released an online version of our algorithm for
testing.
http://blythed.pythonanywhere.com
Details are at:
http://arxiv.org/pdf/1505.01147
Constructive feedback welcome.
Read more:
http://www.letsrun.com/forum/flat_read.php?thread=7287386#ixzz48npcuvUM
Interesting. As an engineer and athletics enthusiast, I have always been fascinated by the mathematical modelling of athletics performance.
I ran some times through and they seem to coincide with a little calculator I developed for the 400/800/1500 based on my observations of performances of male club to elite level athletes.
1.2 x 400m + 0.216 x 1500m = 800m
Of course your work has far more depth, covering a wide range of distances and I will read your paper to see the detail behind your study.
It's simply nice for me to have some confirmation of a rule of thumb I have been using!
It is very impressive to develop an algorithm that describes how athletic ability changes across such a wide range of distances. The differences in physiology between individuals alone is enough to make finding a universal model difficult. Not to mention the unreliable data to hand.
Good work indeed.
Hi Metric Miller,
Thanks for your comments.
Your formula looks good; our work uses the data directly to compute "formulae" of sorts, but which don't admit a simple expression as yours does.
Since your formula was inferred from data, it's good to see that they coincide.
It is encouraging to see methods inferred from different data sets yielding similar predictions.
The 400-1500m are particularly hard to forecast (see the paper) since there are multiple factors at play here: speed, endurance and a third factor which isn't easy to pin down - possibly a lactate resistance factor. That's still open.
I would be interested to hear what you think of the three-number summary (there are two buttons). That's the button on the right. There's a bit of interactive javascript in there, so you can understand the effect of each event on the athlete type.
dgfsgs wrote:
The author is right, however that is assuming the marathon is run on the track in spikes just like his 10000m world record. Kenny would never be able to run 26:17 on the road in flats, so don't expect the marathon and half marathon world records to align with track records.
If the marathon and half marathon were run on the track in spikes, assuming the athletes would be able to finish without loss of muscular elasticity from the higher level of pounding, the marathon world record would immediately be at least a minute faster, and likewise thirty seconds for the half.
I like how you go this route and then don't consider how slow running a turn every 200m for a marathon on a track would be vs a course like Berlin. And yes no one is running a marathon in spikes. I watched Mosop destroy the 25k and 30k wr's on the track and his shoe of choice was a flat for even that distance. So yes, the author could make that reach - should he be right - because both distances are being run in their optimal conditions
Why the aggressiveness? We have been putting up with Ventolin's crazy stuff for a decade maybe, and now we should be complaining about someone making serious predictions?
Duncan Blythe wrote:
Following the recent discussion with Canova and co.
we have released an online version of our algorithm for
testing.
http://blythed.pythonanywhere.comDetails are at:
http://arxiv.org/pdf/1505.01147Constructive feedback welcome.
Read more:
http://www.letsrun.com/forum/flat_read.php?thread=7287386#ixzz48npcuvUM
I have not read your "Prediction and Quantification of Individual Athletic Performance" paper. So I don't know the methods you have used or what exactly you want to predict.
But the Mile WR gives a 3:22.73 for 1500m, which obviously isn't anywhere close to 3:43.13. Also when adding times for other distances (1:43; 12:50) the outcome isn't very helpful at all. So, is it my mistake or my interpretation? Thanks.
No answer?
The Mile WR is not anywhere close to 3:22.73 over 1500m; no one with the potential to run 3:43.13 has any chance to run 3:22.73. But that's what the calculator says.
Have I misinterpreted something?
2:02 flat if he was in peak shape
said88 wrote:
No answer?
The Mile WR is not anywhere close to 3:22.73 over 1500m; no one with the potential to run 3:43.13 has any chance to run 3:22.73. But that's what the calculator says.
Have I misinterpreted something?
I read some of their paper.
The calculator is based upon data of UK athletes. These runners do not race the mile often. Hence it is common to get 3:35 guys like Charlie Grice only run 3:57 as their sole mile attempt. This skews the mile predictions hugely.
If you use the calculator for the metric distances only, from my limited testing it seems to be pretty good.
If you want to use it to get a mile prediction, look at the 1500m prediction and extrapolate yourself with a common factor such as 1.08 or using milesplit.com to get an equivalent performance.
Jakob Ingebrigtsen has a 1989 Ferrari 348 GTB and he's just put in paperwork to upgrade it
Is there a rule against attaching a helium balloon to yourself while running a road race?
How rare is it to run a sub 5 minute mile AND bench press 225?
Am I living in the twilight zone? The Boston Marathon weather was terrible!
Strava thinks the London Marathon times improved 12 minutes last year thanks to supershoes
Mark Coogan says that if you could only do 3 workouts as a 1500m runner you should do these
Move over Mark Coogan, Rojo and John Kellogg share their 3 favorite mile workouts