AI software for the do-it-yourself capper

[Capper Al]

A friend once showed me some Artificial Intelligence software where you put in your data with the results and it figured out your data weights and winning formula. Of course, the bigger the sample size the better the formula would be. It was simple enough to use. Didn't have to be a statistician to use it. Once a sample set was inputted, all one had to do was enter the up coming race and it would also give you a prediction based on your sample. Does anyone know of such software available today? Hopefully, not too expensive.

Thanks

[Hoofless_Wonder]

The first one that comes to mind is the Neurax product on the BRIS site. V3.1 uses AI.

http://www.brisnet.com/php/fw/softwareDisplay/

I've never tried it, and I don't know how much content can be input by the user.

Are you looking for something that's "data independent" from the products like this currently on the market?

[Capper Al]

The first one that comes to mind is the Neurax product on the BRIS site. V3.1 uses AI.

http://www.brisnet.com/php/fw/softwareDisplay/

I've never tried it, and I don't know how much content can be input by the user.

Are you looking for something that's "data independent" from the products like this currently on the market?

The difference between what Neurax does and what I'm looking for is as follows:

Neurax utilizes BRIS data with their own statistical formula and then makes their projections from that combination of the coming race's data and formula. What I need is an app that takes my data and gives me my own formula and allows me to project my picks based on the formula it created from analysis of my data.

"Are you looking for something that's "data independent" from the products like this currently on the market?"

Yes


Thanks

[DeltaLover]

figured out your data weights and winning formula.
Thanks

The real challenge following this approach is not the AI component but way the data are going to be presented to the algorithm...

In an ideal world an AI agent should be able to receive the data in their most purest format meaning a complete data base with race charts and provide as output the winning percentages of each horse.

In the real worlds though, we can not be so pure. Abstractions like track variant, speed figures, power ratings, pace figures or even the final odds(!) etc are going to be needed for the algorithm to produce some reasonably accurate results which means that again the solution to the problem is referred to the set of figures to be used.... The good part of this approach is that it makes it relatively easy to justify the effectiveness of a any figure, handicapping factor (TRUE-FALSE), rating or even raw data... The not so good part of it is that it consists of a very time consuming sequence of trial and error assumptions... (I hope the length of this sequence does not extend to the magnitude of a human life span)

[hcap]

I have tried a few. I don't think they are applicable to horse racing. They do an excellent job red boarding leading one to believe what APPEARED to have worked historically, will hoid going forward. I have tried short term modeling, a few days (down to even a few races), and long term, up to 12 month models. Modeling in Excel using simply win percent or impact values of discrete home made factors seemed to work as well as neural nets, if not better going forward.

Best approach I have found is dynamic modelling. I choose my factors, build a model based on say 30 racing days. Handicap day number 31. Record how well cycle number 1 performed. Advance the model by dropping racing day number 1 and adding day 31. Handicap day 32. Let it run for 300 cycles. Come back and see how my configuration of models, choice of factors, and the number of racing days in each cycle has done over an extended period.

For starters

http://www.google.com/search?hl=en&source=hp&biw=&bih=&q=neural+networks&oq=neural+networks&aq=f&aqi=g10&aql=&gs_sm=3&gs_upl=2185l9909l0l16887l15l15l0l4l4l0l416l1321l9. 1.4-1l11l0#hl=en&sclient=psy-ab&q=neural+network+softwre+&oq=neural+network+softwre+&aq=f&aqi=g-l4&aql=&gs_sm=3&gs_upl=0l0l1l227l0l0l0l0l0l0l0l0ll0l0&gs_l=serp.3..0i13l4.0l0l1l227l0l0l0l0l0l0l0l0ll0l0&pbx=1&bav=on.2,or.r_gc.r_pw.r_qf.,cf.osb&fp=d4ad57653d4016cc&biw=1070&bih=731

[LemonDropKid]

I have used Neuroshell2. They even have a racetrack add-in but I did my own data prep. Like anything else, you have to tinker with it to get the best results. More data is better.

http://www.wardsystems.com/neuroshell2.asp

[Capper Al]

I have used Neuroshell2. They even have a racetrack add-in but I did my own data prep. Like anything else, you have to tinker with it to get the best results. More data is better.

http://www.wardsystems.com/neuroshell2.asp

It looks interesting, but I'm not seeing anything under $800.00.

[Dave Schwartz]

If you are interested in Neural Nets, try this one:

http://www.justnn.com/

It is free!

[InControlX]

Ditto to hcap's experience... Some pattern "finds" looked great in the samples but never panned out in subsequent runs. One problem I noticed with the simpler neural approaches is a lack of reasonable precision limits to the data, instead using a fixed decimal assignment to everything. This creates artificial hits which will never repeat.

ICX

[Dave Schwartz]

I used to write neural networks. There is an inherent problem with NNs and horse racing.

The problem is that the goal of a NN is to create a solution that encompasses every race within a certain threshold.

The problem is that there is no place where it says, "This race cannot be successfully handicapped."

It assumes that every answer is getable; every race makes sense.

This forces the handicapper to make sure that every race that is handed to the NN is, in fact, logical. Otherwise, the NN finds a way to make them all fit.

"Finding a way" includes rules that cause horses to win because they had too little of something - like (say) speed rating in the last race or a bad jockey. Those are not valid reasons for a horse losing any more than saying a horse lost because his jockey was too good.

In a nutshell, NNs want to get 100% of the races like 90% right. What we need is a system that gets 90% of the races 100% right.


I suggest you guys look into genetic algorithms. They are much better suited for horse racing.


Regards,
Dave Schwartz

[Capper Al]

I used to write neural networks. There is an inherent problem with NNs and horse racing.

The problem is that the goal of a NN is to create a solution that encompasses every race within a certain threshold.

The problem is that there is no place where it says, "This race cannot be successfully handicapped."

It assumes that every answer is getable; every race makes sense.

This forces the handicapper to make sure that every race that is handed to the NN is, in fact, logical. Otherwise, the NN finds a way to make them all fit.

"Finding a way" includes rules that cause horses to win because they had too little of something - like (say) speed rating in the last race or a bad jockey. Those are not valid reasons for a horse losing any more than saying a horse lost because his jockey was too good.

In a nutshell, NNs want to get 100% of the races like 90% right. What we need is a system that gets 90% of the races 100% right.


I suggest you guys look into genetic algorithms. They are much better suited for horse racing.


Regards,
Dave Schwartz

Thanks

[bob60566]

I used to write neural networks. There is an inherent problem with NNs and horse racing.

The problem is that the goal of a NN is to create a solution that encompasses every race within a certain threshold.

The problem is that there is no place where it says, "This race cannot be successfully handicapped."

It assumes that every answer is getable; every race makes sense.

This forces the handicapper to make sure that every race that is handed to the NN is, in fact, logical. Otherwise, the NN finds a way to make them all fit.

"Finding a way" includes rules that cause horses to win because they had too little of something - like (say) speed rating in the last race or a bad jockey. Those are not valid reasons for a horse losing any more than saying a horse lost because his jockey was too good.

In a nutshell, NNs want to get 100% of the races like 90% right. What we need is a system that gets 90% of the races 100% right.


I suggest you guys look into genetic algorithms. They are much better suited for horse racing.


Regards,
Dave Schwartz

Can Algorithms predict over the short or long term.

Mac:confused: