Automatic Training Plan Generation

23 watchers
Dec 2012
2:50pm, 10 Dec 2012
12,124 posts
  •  
  • 0
fetcheveryone
(the reason why I supply it with a 1:20-1:50 bracket is just to limit the number of cases it looks at.
Dec 2012
2:53pm, 10 Dec 2012
10,743 posts
  •  
  • 0
Naomi P
Ok. Interesting. Biggest differences are I've done more halves than your matching case and fewer miles. The rest of it, apart from the outcome and conversion, are very similar. Looks like I didn't bust a gut in the race. *cough*

That outcome is a lot closer to my 1/2 PB pace (1:35:02) though. And my training probably wouldn't be that different to what it is now, I'd just make the speed sessions better quality.
Dec 2012
2:55pm, 10 Dec 2012
12,125 posts
  •  
  • 0
fetcheveryone
The reason why I'm doing all the guessing, is to try to prove the concept i.e. once the guesses start to get more accurate, I can play it backwards e.g. if you said to me you wanted 1:29, I could find appropriate matches, and spit out the training schedule used by that person.
Dec 2012
2:56pm, 10 Dec 2012
10,744 posts
  •  
  • 0
Naomi P
Sure, I thought so. I like the methodology :-)
Dec 2012
2:57pm, 10 Dec 2012
12,126 posts
  •  
  • 0
fetcheveryone
At the moment, I'm measuring the similarity of cases based on the following:

total mileage
days per week
longest run
longest five runs
average training pace
number of speedwork sessions

Suggestions for other comparisons are welcome :-)
Dec 2012
2:59pm, 10 Dec 2012
14,370 posts
  •  
  • 0
JohnnyO
age?

I just read that CBR link. There are parallels with the scoring systems we use in medicine (to predict outcome etc.) Its very interesting.
Dec 2012
3:02pm, 10 Dec 2012
10,745 posts
  •  
  • 0
Naomi P
How about difference between speedwork pace and average training pace?

Not easy because I guess a lot of people (inc. me) log the time for the whole run, rather than the interval / tempo stuff etc. But there's often a big spread between average training pace and the pace at which you do your hard stuff.
Dec 2012
3:02pm, 10 Dec 2012
12,127 posts
  •  
  • 0
fetcheveryone
There's a bunch of medical-based CBR systems - it's a common example used to explain how it works i.e. patient exhibits symptoms, doctor recalls other cases with similar symptoms, some are dismissed based on contraindications, a diagnosis is made based on nearest matching case, treatment follows that case.
Dec 2012
3:03pm, 10 Dec 2012
14,372 posts
  •  
  • 0
JohnnyO
Never thought of it like that, but you are absolutely right.
Dec 2012
3:08pm, 10 Dec 2012
14,373 posts
  •  
  • 0
JohnnyO
I think the biggest problem is standardising all the inputs. Like Naomi says, what I call speedwork differs from others, and people record it differently too. Some people stop the watch for the rest periods etc.

Not all speedwork is equal, so whilst 'number of sessions' is a very pragmatic way of factoring it in, there may be useful data that is lost.

Without reading back, are you attempting to match individual cases (who ere succesful), or match your desired outcome to a population who were close? I suppose a lot of the variations in terminology will even out with a bigger sample size.

About This Thread

Maintained by fetcheveryone
I'm messing around with some data, and trying to generate training plans using case-based reasoning.

TREAT ANY RESULTS WITH EXTREME CAUTION!!!!!!!!!!!!

Related Threads

  • training









Back To Top
X

Free training & racing tools for runners, cyclists, swimmers & walkers.

Fetcheveryone lets you analyse your training, find races, plot routes, chat in our forum, get advice, play games - and more! Nothing is behind a paywall, and it'll stay that way thanks to our awesome community!
Get Started
Click here to join 112,111 Fetchies!
Already a Fetchie? Sign in here