|
Dec 2012
3:19pm, 10 Dec 2012
12,128 posts
|
fetcheveryone
Absolutely - speedwork is the big variable here. Personally when I do intervals, I record from the start of the first effort to the end of the last effort, and all the breaks in between - all in one chunk. I feel that if I stop the watch during the breaks, I'd get an unrepresentative recording of how fast I was able to cover the distance of the efforts. But I know of people who record each effort as an individual session. I've looked at recording the number of minutes of speedwork, rather than the number of sessions, but again, the results are variable.
One way that might work better, but at quite a high computational cost, would be to analyse the detailed plots from GPS data - so you could see exactly how much running at pace the user had done.
Not sure I quite understand your last question. The guesses I pulled out for Naomi were the five 'nearest' cases. I could either examine the closest cases and discard the weird ones, or do some sort of amalgam of them all.
|
|
Dec 2012
4:27pm, 10 Dec 2012
12,129 posts
|
fetcheveryone
Naomi - here's some interesting stuff - here are your conversion rates for half marathons to date:
RACE DATE = 2008-02-03 RACE TIME = 01:51:10 RACE PACE = 509 TRAINING PACE = 555 RATIO = 0.918
RACE DATE = 2008-10-05 RACE TIME = 01:48:51 RACE PACE = 499 TRAINING PACE = 545 RATIO = 0.915
RACE DATE = 2009-02-01 RACE TIME = 01:44:44 RACE PACE = 480 TRAINING PACE = 534 RATIO = 0.899
RACE DATE = 2011-02-06 RACE TIME = 01:47:33 RACE PACE = 493 TRAINING PACE = 534 RATIO = 0.923
RACE DATE = 2011-02-13 RACE TIME = 01:38:39 RACE PACE = 452 TRAINING PACE = 529 RATIO = 0.853
RACE DATE = 2011-06-12 RACE TIME = 01:41:59 RACE PACE = 467 TRAINING PACE = 507 RATIO = 0.921
RACE DATE = 2012-02-19 RACE TIME = 01:35:02 RACE PACE = 435 TRAINING PACE = 495 RATIO = 0.879
RACE DATE = 2012-03-04 RACE TIME = 01:46:23 RACE PACE = 487 TRAINING PACE = 498 RATIO = 0.978
RACE DATE = 2012-07-15 RACE TIME = 01:43:16 RACE PACE = 473 TRAINING PACE = 490 RATIO = 0.964
RACE DATE = 2012-08-19 RACE TIME = 01:47:07 RACE PACE = 491 TRAINING PACE = 491 RATIO = 0.999
RACE DATE = 2012-09-30 RACE TIME = 01:39:12 RACE PACE = 454 TRAINING PACE = 497 RATIO = 0.914
RACE DATE = 2012-11-04 RACE TIME = 01:37:07 RACE PACE = 445 TRAINING PACE = 504 RATIO = 0.882
So your average training pace has crept down from about 540s/mile for your first five halves, to around the 500s/mile region for the most recent seven. Converting that to race pace has gone as low as 88%. I think that if I propose a conversion rate of 88%, and weigh the cases against that too, it would have a good impact.
|
|
Dec 2012
4:34pm, 10 Dec 2012
14,375 posts
|
JohnnyO
I think you answered it. To give Naomi a training plan you find one person who was successful and whose actual activity matches what she is willing to do.
What we do is take a population and measure a load of things that we think might be important (age, blood pressure, speed sessions, total mileage etc) and outcomes (death, recovery, achieving a certain target time). We then use logistic regression analysis (which I don't really understand) to find those factors which actually affected the outcome.
Tise factors which do affect outcome are then scored/weighted and put into a model which is then used prospectively to attempt to predict outcomes. The system can be re-weighted if necessary and run again.
This can give very accurate results for a population (if 100 patients turn up with these characteristics then 20% of them will die), but not for individuals (it doesn't mean you have an 80% chance of survival- your individual chance is still either 100% or 0%, you cant be 80% alive).
Anyway, I am rambling, but essentially my point is that we use a group of people with similar characteristics to predict outcome. You are just using one to generate a plan to affect outcome. Is there enough data to find a pool of people that nearly match Naomi, and average their activity?
|
|
Dec 2012
4:52pm, 10 Dec 2012
10,746 posts
|
Naomi P
The question is, are there many people like me? Surely not 
That's very interesting data in your post Fetch, thank you. I'll definitely be having a closer look at that and having a look back through the thread at other people's conversion ratios if you've posted them.
I like the way you use seconds / mile too. Makes it so much easier to compare.
|
|
Dec 2012
5:05pm, 10 Dec 2012
12,130 posts
|
fetcheveryone
Not necessarily just one person - but in essence, yes, the best fit.
The nearest neighbour model in CBR gives each case parameter a weight to show its relative importance. At the moment, I'm just playing with those, but I guess I could do more to determine the reliability of each factor in determining the outcome.
My economics teacher in school used to refer to the quote: "man has a mother, but mankind does not" - in other words, one man and a population of men are not the same thing. And it's the whole probability vs possibility thing, my turn to ramble now - must go re-read my Bart Kosko
Hoping there's enough data - I have eight years worth
|
|
Dec 2012
5:09pm, 10 Dec 2012
12,131 posts
|
fetcheveryone
Tea time here, catch you all in the funny pages.
|
|
Dec 2012
5:33pm, 10 Dec 2012
5,324 posts
|
goldfleecyandmyrrh
Are you basically doing Structural Equation Modelling then? Have a load of variables, fiddle with weightings for each till you get the best match and come up with an equation which best fits the data?
We did a bit of this at uni, but the software was shit so nobody could get it to work! The concept seems simple enough though, although as previously said because not everyone records their data in a standardised way you lose some accuracy there. For most variables you would have a standardised measure. Could you just use number of interval sessions per week plus average interval pace perhaps? Having different length intervals makes it tricky too, I suppose.
|
|
Dec 2012
6:25pm, 10 Dec 2012
138 posts
|
colettedeann
thanks fetch - i will get that in on my long runs in the next few weeks
|
|
Dec 2016
8:33am, 5 Dec 2016
8 posts
|
running_dan_w
As a coach and a data geek this is very interesting. I like the concept although are you able to include (or model) other variables that are likely to be significant like training background - how many years they've run, how many runs or miles per week they've been running?
|
|
Jun 2019
4:45pm, 14 Jun 2019
12,788 posts
|
Rosehip
Bounced
Boing
something a mix of project joker and a weekly mileage calculator might be interesting
|
