Horse Racing Forum - PaceAdvantage.Com - Horse Racing Message Board

Go Back   Horse Racing Forum - PaceAdvantage.Com - Horse Racing Message Board


Horse Racing Forum - PaceAdvantage.Com - Horse Racing Message Board

Horse Racing Forum - PaceAdvantage.Com - Horse Racing Message Board (http://www.paceadvantage.com/forum/index.php)
-   Handicapping Software (http://www.paceadvantage.com/forum/forumdisplay.php?f=3)
-   -   Neural Networks (http://www.paceadvantage.com/forum/showthread.php?t=54073)

raybo 01-17-2009 11:58 AM

Neural Networks
 
I've always been interested in NNs for horse racing since I first heard of them several years ago. I realize that many have used them with varying degrees of success or failure. I also realize that many prefer regression analysis over NNs, however, while being very good at math and logic, I am not highly educated in the higher forms of math or programming, which would be required to get into regression based work.

I have on my computer a couple of spreadsheet based NN programs that I downloaded off the internet, free of course, and have messed around with them some. I did create a NN for drag racing (bracket racing) as I crew for a friend who runs a gas dragster. This NN is very basic, having only 5 inputs and 2 outputs. The key to this one is being able to predict your next ET (elapsed time) as you must post what you think your ET will be prior to the race, which is then used by the timer to program the "Christmas tree" lights, thereby spotting the competitor or receiving a spot from the competitor. This little program works surprisingly well. However, this NN only has to train on 1 car, not even caring about the capabilities of the other competitor in the race. So you can train it using data from previous runs and simply have it predict the next ET for the same car.

Horseracing is a different story, as you can imagine, more inputs are involved for each horse and there are several horses involved in each race and there will be, many times, more than 1 horse that can affect how the race is run (pace of the race). Also, I have not figured out how one would be able to input today's prerace data; distance, post positions, weights, etc., into the NN, or if that data is even needed.

The output(s) would depend on what one wants to predict; final times, pace times, speed figures, finish positions, final B/Ls, etc.. It would be nice to be able to predict all the entries' finish positions or final B/Ls, but that would require having multiple outputs. Would you need to run the predictor for each horse in order to predict the desired output for each horse in the race? Or could you input all the horses' data and get an output for each horse?

Dave Schwartz 01-17-2009 02:21 PM

As someone who has written many neural nets from scratch in the past 20 years, I can tell you that there is an inherent problem with neural nets.

There are three problems with NN's.

The first problem is that the nature of a NN is that it wants to be perfect and it is very sensitive to "bad" data.

What is "bad" data? Well, an obvious example of bad data would be a race where (say) the #1 horse turned right coming out of the gate and wiped out 3 other horses allowing a $168 winner to romp home.

When the NN trains it assumes, logically, that in every race you gave it the winner should be able to become the best horse. In the case outlined above, the brain will do anything it can to "jusitfy" that $168 winner.

It will even say that the reason this horse won the race is that he was ranked last in the field for (say) FT in the last race. This is not a reason for winning but, by changing around all of its weighting system it can get this horse (which is what it will do).

Now, one can put in the time to remove the more obvious races... those with DQ's. One should not jsut remove the large price-races - that would be folly. The real problem with this is that there are simply "atypical" races where the absolute, dead-bang, dead-nuts, can't-lose horse loses. Why did he lose? Not because of anything that was predictable by looking at the PPs.

One such horse that comes to mind was Bayakoa at Del Mar many years ago. As I recall, it was a 6 horse field and Bayakoa, a 1/5 or 2/5 favorite (who looked worth that price) finished out of the money. The horse was simply first for everything. How does the brain reconcile that sometimes the 1/20 horse loses?

The answr is it can't. - which brings us to the 2nd problem.

With a NN, there is a threshold score (between 0.00 and 1.00) for an event happening. Take a basketball game. This is simplistic, but the way it typically works is that if the score is above 0.90 you bet the home team and below 0.10 you bet the visitor (or favorite/dog, whatever).

In other words, there is a 10% "margin for error." Now recall that the system is trying to get every "fact" right. Essentially, its goal is to get 100% of the games/races 90% right.

What we need is a system that tries to get 90% of the races 100% right. In other words, it needs to be able to say, "Hey, there are just some races that I can't get."

Finally, there is a third, though less important drawback. The system becomes good at predicting what it gets the most of. If you look at 100 races, about 800 horses, you will have 100 winners and 700 losers. A NN is much better at predicting losers than it is winner because that is where its experience lies.


GAs are a better deal.

Regards,
Dave Schwartz

Greyfox 01-17-2009 03:41 PM

Quote:

Originally Posted by Dave Schwartz
GAs are a better deal.

Okay, I'll look dumb here. What's a GA?

raybo 01-17-2009 04:07 PM

Quote:

Originally Posted by Greyfox
Okay, I'll look dumb here. What's a GA?

Stands for generic algorithm. Pretty complicated, much more-so than I can handle, anyway.

Dave Schwartz 01-17-2009 05:46 PM

Raybo,

Just like neural nets, there are commercial GAs out there and even some free ones I have heard.



Dave

raybo 01-17-2009 06:01 PM

Quote:

Originally Posted by Dave Schwartz
Raybo,

Just like neural nets, there are commercial GAs out there and even some free ones I have heard.



Dave

Thanks Dave. I'll do some looking around.

Dave Schwartz 01-17-2009 08:53 PM


dartman51 01-17-2009 11:21 PM

Back in the early 80's, I bought a program from RACE COM in Ormond Beach, FL. It was a NN. It worked so so for a while, then stunk up the place. I have never heard NN's explained. Great job Dave, now I understand why I had such a hard time with it. Thanks for the explanation.:ThmbUp:

zerosky 01-19-2009 04:32 PM

Quote:

Originally Posted by Dave Schwartz
With a NN, there is a threshold score (between 0.00 and 1.00) for an event happening. Take a basketball game. This is simplistic, but the way it typically works is that if the score is above 0.90 you bet the home team and below 0.10 you bet the visitor (or favorite/dog, whatever).

In other words, there is a 10% "margin for error." Now recall that the system is trying to get every "fact" right. Essentially, its goal is to get 100% of the games/races 90% right.

What we need is a system that tries to get 90% of the races 100% right. In other words, it needs to be able to say, "Hey, there are just some races that I can't get.


Excellent insight as usual, your posts are much appreciated. In my brief foray into AI I ended up trying to decipher the Dempster-Shafer Theory.

The probability contains two components represented by Belief and Plausibility. As per usual I got hopelessly lost in the Math.


http://en.wikipedia.org/wiki/Dempster-Shafer_theory

http://www.glennshafer.com/

raybo 01-19-2009 05:11 PM

Quote:

Originally Posted by Dave Schwartz

Dang! I thought NNs were complicated. GAs are totally out of reach for my, otherwise above normal IQ. The math just throws me for a loop so bad that I just turn my computer off and try to completely blank my mind.

This stuff will, literally, drive you nuts!

Dave Schwartz 01-19-2009 06:49 PM

Raybo,

You are working so hard for this, I am thinking I should give you an hour of consult to clear your head.

If you're interested, drop me a PM and we'll work out a time.


Dave


All times are GMT -4. The time now is 07:40 PM.

Powered by vBulletin® Version 3.8.9
Copyright ©2000 - 2024, vBulletin Solutions, Inc.
Copyright 1999 - 2023 -- PaceAdvantage.Com -- All Rights Reserved

» Advertisement
Powered by vBadvanced CMPS v3.2.3

All times are GMT -4. The time now is 07:40 PM.


Powered by vBulletin® Version 3.8.9
Copyright ©2000 - 2024, vBulletin Solutions, Inc.
Copyright 1999 - 2023 -- PaceAdvantage.Com -- All Rights Reserved
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program
designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.