artificial intelligence

[acorn54]

was just on the tradecision site. it's software that is used in buying -selling equities ,futures and currency.

http://www.tradecision.com/index.htm

it seems that they are using artificial intelligence to make money.
if you look at the purchase price there are two versions of the software. the one with the artificial intelligence incorporated into the program is almost double the price,so it seems this is a very valuable tool.
hsh, i think has an artificial intelligence application. if any hsh users would like to share there experiences and successes with the group i'm sure there are some that would be most interesed. i know i would.
of course if you are successfully using artificial intelligence in the application of handicapping i would not be surprised if there are no responses
guy (acorn)

[MarylandPaul@HSH]

Acorn, HSH has two distinct AI modules. The first is the "Ants", which is, to my limited understanding, more of a neural network approach. I don't have a lot of experience using it, but I do know that it's performance is sensitive to what you feed it. For example, if you give it "ungettable" races, ie races in which the winner doesn't show any positives in the program's core factors, the resulting network may not be as reliable. I do know of several users who have been using Ants for quite some time, so I'd assume they like what they're getting out of it. (Hi Tony! :)

The other is the "Swarms" module. It's more experimental in nature. Using swarm intelligence, thousands of elements (virtual horseplayers, if you will..) are created, given randomly chosen and weighed factors, and left to fight it out in their own win pool. The survivors, having made more money, move to the top of the list, while the losers either move to the bottom, or are elminated and replaced with new elements (your choice). Once training has completed, (completion criteria is your choice also) you're left with a listing of the most profitable elements, and the factors and weightings they used. You can then view an odds line based either on the single best element, or a weighted-by-profit consensus of the entire list.

I've been testing and using the Swarms module fairly extensively. IMO, a fair amount of thought has to go into the setup to get the most out of it. You have to choose the filtered subset of races to run the swarm against, the factors to use (and how many each element uses), how they're weighted, when the training should end, how many horses each element will play, and more.

An approach I'm testing right now uses two filters for each race. One is pace specific, the other uses non-pace factors (both filters dynamically created for the current race). I then run a single pass through 20000 randomly created elements for each filter, then view an odds line created from both swarms. I don't have hard data yet, as I'm still tweaking, but so far it's doing an outstanding job of getting the winner in the top 4 in races with 10+ fields.

That's a pretty sketchy description; any questions, just yell.

MP