Face Lift

No new functionality for the latest version of FSGL Resultats, but a complete new look with a colour scheme that is a bit easier on the eye, some icons, better allignment etc. A refresh maybe necessary to get the new stylesheet.

Ekiden

Here's what the form guide predicts for the Ekiden teams. A '*' means there was insufficient data to predict and so the values are best guess.

Distance	Vet1	Time
5	Fab	00:18:35
10	Vincent *	0:37:44
5	Bruno B	00:18:38
10	Thiery Maz	0:37:44
5	Charles	00:18:53
7.195	James	00:26:28
Total		02:38:02

Actual Time  = 02:39:38

Distance	SH1		Time
5		Mohamed Y *	00:16:53
10		Anis	00:34:37
5		Nico	00:17:39
10		Mohamed S	0:32:44
5		Jérémie	00:16:53
7.195		Eric	00:24:30
Total			02:23:16

Actual Time = 02:23:57

Distance	SH2	Time
5	Gerald	00:18:04
10	Robin	00:38:28
5	Simon	00:18:50
10	Fred *	00:38:28
5	Nick B	00:18:47
7.195	Romain	00:25:10
Total		2:37:47

Actual Time = 02:39:11

Distance	SF1	Time
5	Nadia	00:22:12
10	Françoise	00:43:23
5	Catherine	00:21:48
10	Mirelle	00:36:19
5	Aude	00:21:43
7.195	Delphine	00:32:05
Total		2:57:30

Actual Time = 02:59:13

Distance	VH2	Time
5	Daniel	00:20:48
10	Adrian	00:42:04
5	Patrick	00:19:48
10	Alex	00:40:22
5	Gerard	00:20:11
7.195	Frank	00:30:16
Total		2:53:29

Actual Time = 02:52:31

How Hard was that Race? New Features in v1.1

Race Ratings

“Courses en 78” is a new feature that allows you to compare the difficulty level of each race in the Yvelines. Based on the 2010 performances of the top 50 athletes from the Challenge 78, each race is listed, grouped by distance, with a corresponding percentage difficulty rating. A reference time is also included to make it easier to see the how the difficulty of each race affects performance. Both the reference time and the reference race are configurable.

The difficulty table is built by first assigning each athlete in the sample a performance rating for the 2010 season using the same algorithm as used for the Prediction feature. The rating is then compared against the actual performance of each athlete in the races they competed in and the difficulty table is built accordingly. Any performance that differs by more than 5% from an athlete’s average is deemed to be an anomaly and not included in the calculation. A race needs at least four qualifying athletes before it is included in the table.

Comparing my own performances from 2010 over 10 km and using La Fin D’Oisienne as the base reference time, the following results area obtained.

Race	Actual Time	Predicted Time
LA FIN D'OISIENNE	00:39:17	00:39:17
FOULEE D'AUBERGENVILLE	00:38:44	00:38:45
TOUR DE HOUILLES	00:39:15	00:39:26
CHALLENGE CHRISTIAN LAGOUTTE	00:40:00	00:39:55

The values for actual and predicted times are astonishingly close. Assuming that the difficulty ratings are at least partially accurate, then the results suggest that my form was fairly constant throughout the season whereas I actually perceived it as being much more up and down.

Of course, this kind of calculation is open to all sorts of external influences, especially when trying to use it as a guide for future races. Most obviously, climatic conditions on the day of a race can significantly influence athlete performance. It is perhaps most useful as a historical view of an athlete’s form throughout the 2010 season although still valid for getting a general idea of the difficulty level of a future race.

See the new feature here: Race Ratings

Ver 1.0

This is the first official release of the results site and is now finally (more or less) in French. There are only a few visual changes from 0.6 as well as a couple of bug fixes.

Ver 0.6 is here.

The main change in ver 0.6 is a rewritten prediction page that perhaps should really be called form guide. Instead of listing a single set of times for each distance, 5 metrics are now included. Most are self explanatory, but the “Adjusted Form” metric requires a bit of explanation. As with the standard Form metric, recent races are analyzed to produce a set of equivalent times for each distance. Before including a result in the final calculation, it is compared with an athlete’s average equivalent performance over the past 12 months. If it differs positively by more than 5% it is rejected from the final analysis. This will hopefully eliminate races which were particularly difficult, e.g. due to a large hill, or where the athlete was running at less than maximum capacity, resulting in a more accurate analysis.

Which metric is the most accurate? As ever there is not a simple answer as performance on the day can depend on many external factors that cannot be incorporated into the prediction algorithm. It’s probably best to consider the metrics together, remembering that the results will be more accurate if the athlete has raced frequently in the preceding months.

Other new features include a VMA calculator. Enter your reference time over 1500m or 2000m and the calculator will generated interval times over various distances for VMA training sessions. There are also a few bug fixes and some new help menus that make use of some Ajax collapsible panels. They are somewhat basic at the moment but should get better with time.

One feature that didn’t make the cut was another prediction metric base on performance trend over recent races. The idea is to use Linear Regression over past results to produce a trend of whether an athlete’s form is improving or not. In theory it should then be possible to predict a range of expected times for an athlete’s next race based on confidence intervals on the previously produced line. Not surprisingly, results can sometimes be fairly wild as the majority of athletes do not produce enough reference data to allow for a precise analysis. I will probably end up restricting this future metric to athletes who have raced at least 4 times in the past 6 months.

Tour de Houilles

My predictions for a few of the athletes who are running the Tour de Houilles this Sunday are below, as usual, calculated with my prediction tool found here.  They look reasonably accurate to me although I hope Nico can go under 37 minutes given his recent form in the French Cross Country Championships, which the prediction tool can’t take into account.

Nick Bramwell	0:38:31
Eric Chenallier	00:35:21
Jean-Marc Chretien	00:35:14
Gerard Diolez	00:42:23
Patrick Le Gouguec	00:42:09
Nicolas Rousseau	00:37:31
Gerald Dejean	00:36:30

This is a race I usually like to do myself, normally finishing in a little over 39 minutes but not this year due to injury.  There is no sign I’m going to restart training any time soon which is fine for now, but might be annoying as the weather gets better.  It might be a good idea to finally buy that road bike I always claim I’m going to get.

 If anyone wants to predict their times for the above or others from the club it would be good to see who is the most accurate.

Version 0.5 - Improved Stats and a Better Prediction Tool

Version 0.5 is now released with the following new features and enhancements.

·         Improvements to the Race Predictor Tool to include a 5km estimate.

·         A rewrite of the Best Performances page to allow search by category, distance, sex and year.

·         A change to the predictor tool algorithm to produce more accurate results.

·         A season review (2010) table on the prediction page. I.e. Theoretical average times based on results over a whole season.

·         Some format changes to make the site display better in Chrome and Firefox.

·         Several bug fixes to the results page including the sort order of races when viewing per athlete.

·         An updated athlete list.

2010 Form Indicator

As a natural extension to the race prediction tool, http://fsgl.wni-sec.com/RacePredictions.aspx, I thought it would be good to be able to produce a figure that gives an indication of an athlete’s performance over a whole season. As a starting point, I decided on producing a theoretical average time for 10km based on a year’s results for distances 10km, 15km, 21.1km and marathon. Hence using the same algorithm as for the predictor tool but considering races for a whole season instead of just recent form, the following figures came up for some of the regular runners from the club. Also included is the 10 km figure from the race predictor tool.

	A	B	C
1	Athlete	Predicted 10 km time based on recent form	2010 form indicator (10km)
2	Jean Marc	35.14	35.21
3	James	37.33	38.12
4	Nico	37.31	37.55
5	Gerald	36.37	37.54
6	Mirelle	37.21	37.41

For a first attempt, I’m quite pleased with the results as they are not a million miles away from what I would have guessed. The next race between James and Nico should be interesting, although Gerald should still be way out in front. I’m looking forward to the road season starting again so I can test and refine the algorithm.

One obvious future problem is that if someone jogs around a race, maybe to accompany a slower runner, the statistics will be distorted. A solution could be to exclude results that differ from an athlete’s average performance by more than 2 or 3 standard deviations. This is not too complicated to achieve but means I’ll have to start refining the SQL to ensure query speed is not affected.

Prediction Tool

The site now has a race prediction tool. It is similar to the McMillan Running Calculator except that instead of entering a reference time as an initial input value, it looks at recent performances for each athlete and makes estimates accordingly. For obvious reasons, only road races over 10, 15 and 21.1 km are considered for the initial input value. I suppose I should also include Marathon performances and will do in a future version.

There is also a confidence indicator as to the accuracy of results. The more recent the last few races of each athlete, the higher the confidence level of an accurate prediction. Given the lack of road races at the moment, it’s unlikely that too many predictions will have a high confidence level.

The ratios used for making predictions, once a base performance level has been established, are based on McMillan’s values but it is intended to also have a set of ratios based on performances from within the club. This is not expected to produce significantly better results as not enough athletes compete regularly over the full range of distances in order to produce a large enough sample. If there were, it would also be possible to provide a difficulty level for each race and hence make even more accurate predictions.

Of course the tool is only a guide and can’t know about external factors such as weather, injuries, hangovers etc. If the predictions are wildly inaccurate please comment below and I’ll check the algorithm.