|
06-05-2023, 05:35 PM
|
#1
|
Registered User
Join Date: May 2001
Location: Atlanta, Ga
Posts: 541
|
AI Handicapping
Given the current interest in artificial intelligence, I would like to get some feed back from AI handicappers who frequent this forum.
1. How large are your training and test sets (Approximate number of Races)
2. Number of Traditional Handicappers Factors that you use.
3 Have you written code that allows your AI to generate factors form
Traditional Factors
4. If you have AI Generated Factors in your models, what are the Fitness
Function Criteria that you employ for those factors to be included in your
models
5. What Training Methods do employ.
6. Do you work as a Individual Handicapper or on a Team
7. What are your Prediction Targets (Finishing Position, Lengths Beaten, Speed
Rating or other)
8. Typical Training time to achieve a working model
9. Are your predictions Hybrid (Black Box and traditional methods) or Pure
Black Boxes
|
|
|
06-05-2023, 06:30 PM
|
#2
|
Join Date: Mar 2001
Location: Reno, NV
Posts: 16,926
|
Those are learned questions.
Clearly, you are not a novice at this.
My newest AI could best be described as a Deep Learning, Topological Data Analysis, Genetic Algorithm.
The engine is modeled after the LS-1 engine, which ultimately became known as the Pittsburgh Model.
The topological system is all about segmenting the handicapping scenarios.
It has a multi-step process, which includes:
1) Intelligent Data collection
Emphasis on Data shapes - to aid with a convolutional approach.
IOW, the engine builds its own factor combinations.
It also learns via a "top-to-bottom" approach, much as a convolutional DL operation would.
Think: Before you can push to find the most profitable, highest Opt%, etc. the engine must figure out:
* Who the public will likely bet
* Who is likely to go to the front
* Who is likely to have the high probabilities
etc. etc.
Hence, top-to-bottom.
(Technically, I think of it as left-to-right because I visualize the layers as being vertical.)
2) Classification
Which of the hundreds of system components should be used in this race and/or with this horse?
3) Segmentation
This is the topological component. It uses the Classification to decide which segmented system(s) are appropriate for this situation.
4) Handicapping Analysis
Ultimately, it comes down to: A. What are the probabilities.
B. What should the odds be?
5) Betting Analysis
This is steps 1 thru 4, applied to what happens AFTER the race is handicapped. (Logically, these would be steps 5.1 thru 5.4.)
Right now, I am in the middle of coding step 5.
Dave
|
|
|
06-07-2023, 03:34 PM
|
#3
|
Registered User
Join Date: Oct 2022
Location: Southern CA and Las Vegas
Posts: 105
|
Dave is about the best of the best handicappers
around. I think that his approach to AI will
generate positive results.
However, I don't think that AI will make major
strides in handicapping. I do think that
Apple's Vision Pro will. When you can actually
see the upcoming race in three-dimensions,
people will start to identify nuances that
they don't see now.
|
|
|
06-08-2023, 06:42 PM
|
#4
|
Registered User
Join Date: Feb 2009
Location: NY
Posts: 19,026
|
Let me know when AI (or ANY other computer based program) can objectively determine the pre-race mental and physical condition of any animal or the intentions of the connections.
.
.
|
|
|
06-08-2023, 08:46 PM
|
#5
|
crusty old guy
Join Date: Aug 2003
Location: Snarkytown USA
Posts: 3,932
|
And you let us know when the toteboard can do the same, k?
__________________
"Don't believe everything that you read on the Internet." -- Abraham Lincoln
|
|
|
06-08-2023, 09:40 PM
|
#6
|
PA Steward
Join Date: Mar 2001
Location: Del Boca Vista
Posts: 88,737
|
Quote:
Originally Posted by Nitro
Let me know when AI (or ANY other computer based program) can objectively determine the pre-race mental and physical condition of any animal or the intentions of the connections.
.
.
|
is that necessary to win?
|
|
|
06-08-2023, 10:06 PM
|
#7
|
Registered User
Join Date: Feb 2009
Location: NY
Posts: 19,026
|
Quote:
Originally Posted by headhawg
And you let us know when the toteboard can do the same, k?
|
That’s a convenient way to get around answering a question: Just pose another unrelated question!
Besides, you seem to be all over this forum so I’m astonished you’ve missed any of my previous predictions based on the tote info. In case you might not realize it, the intentions very often go hand-in-hand with intimate knowledge the animal’s CURRENT overall condition. It also follows that in many such instances the wagering is a reflection of inside information unbeknownst to the general public.
.
.
|
|
|
06-08-2023, 10:07 PM
|
#8
|
Registered User
Join Date: Feb 2009
Location: NY
Posts: 19,026
|
Quote:
Originally Posted by PaceAdvantage
is that necessary to win?
|
In my book it certainly is!
But maybe your question should be reserved for those who actually believe that AI will become their ultimate answer to solving the horse racing riddle.
.
.
|
|
|
06-09-2023, 07:34 AM
|
#9
|
Registered User
Join Date: Feb 2012
Posts: 2,956
|
AI selections
Stumbled upon this site awhile back Never tracked how their selections do
Most tracks are available for FREE
https://aihorsepicks.com/
|
|
|
06-09-2023, 09:31 AM
|
#10
|
crusty old guy
Join Date: Aug 2003
Location: Snarkytown USA
Posts: 3,932
|
Quote:
Originally Posted by Nitro
That’s a convenient way to get around answering a question: Just pose another unrelated question!.
|
Unrelated? I asked a question in a similar vein as yours.
Quote:
Originally Posted by Nitro
Besides, you seem to be all over this forum so I’m astonished you’ve missed any of my previous predictions based on the tote info.
|
Why astonished? What makes you think that I am interested in your "predictions"? Much like a guy who posts formulas here, your toteboard posts have little value for me. If you want to teach us something specific, fine. Posting predictions is tantamount to making selections. There's a whole section here for predictions; I rarely visit it.
Quote:
Originally Posted by Nitro
In case you might not realize it, the intentions very often go hand-in-hand with intimate knowledge the animal’s CURRENT overall condition. It also follows that in many such instances the wagering is a reflection of inside information unbeknownst to the general public.
|
All connections have intimate knowledge of their horse to a greater or lesser degree. You want the AI to be objective. How about you telling us exactly what percentage of the pools reflect this inside information? Surely you must have that objective number written down somewhere. Otherwise, how would you know if the money in the pools is reflective of insider information or the guy selling tout sheets in the parking lot?
__________________
"Don't believe everything that you read on the Internet." -- Abraham Lincoln
|
|
|
06-09-2023, 12:03 PM
|
#11
|
Registered User
Join Date: Dec 2001
Location: JCapper Platinum: Kind of like Deep Blue... but for horses.
Posts: 5,300
|
Quote:
Originally Posted by Nitro
Let me know when AI (or ANY other computer based program) can objectively determine the pre-race mental and physical condition of any animal or the intentions of the connections.
.
.
|
I'll play. (At least the physical condition/horse physicality part.)
There are sharp people who have developed the ability to recognize fit and ready horses.
They know it when they see it. And it can be a difference maker.
If human beings can learn to do it: AI can learn to do it faster and with fewer mistakes.
Enter image recognition software.
Imagine a team of players, who among other things, record thousands of hours of track video - horses in the paddock, post parades, pre-race warmups, and gate loads, etc. while saving the video in digital format to a hard drive.
From there, have AI train on the individual frames of the video - comparing visible attributes for tens of thousands of horses against odds and finish position.
It's absolutely doable.
Don't assume for a second it isn't being worked on.
-jp
.
__________________
Team JCapper: 2011 PAIHL Regular Season ROI Leader after 15 weeks
www.JCapper.com
Last edited by Jeff P; 06-09-2023 at 12:12 PM.
|
|
|
06-09-2023, 12:47 PM
|
#12
|
Registered User
Join Date: Nov 2012
Posts: 1,559
|
Quote:
Originally Posted by Jeff P
I'll play. (At least the physical condition/horse physicality part.)
There are sharp people who have developed the ability to recognize fit and ready horses.
They know it when they see it. And it can be a difference maker.
If human beings can learn to do it: AI can learn to do it faster and with fewer mistakes.
Enter image recognition software.
Imagine a team of players, who among other things, record thousands of hours of track video - horses in the paddock, post parades, pre-race warmups, and gate loads, etc. while saving the video in digital format to a hard drive.
From there, have AI train on the individual frames of the video - comparing visible attributes for tens of thousands of horses against odds and finish position.
It's absolutely doable.
Don't assume for a second it isn't being worked on.
-jp
.
|
I think the recording of the video part of it for each horse (like so many other good inputs that I keep harping on as being necessary) is what will prove to be elusive. Again I wouldn't underestimate what the computer can do when spoon fed good inputs. The problem with racing data which has always been the case is the value-added pieces aren't readily available in any consumable format, only the more generic information is readily available. The notion that I find entertaining is when people believe AI will generate its own good inputs. Like a world class chef can somehow cook great food with third rate ingredients. Bad meat is bad meat. As someone who has done my fair share of coding it's amusing what people think is possible in this arena.
|
|
|
06-09-2023, 01:39 PM
|
#13
|
Registered User
Join Date: Feb 2010
Location: St. Louis suburb
Posts: 1,764
|
Quote:
Originally Posted by MJC922
I think the recording of the video part of it for each horse (like so many other good inputs that I keep harping on as being necessary) is what will prove to be elusive. Again I wouldn't underestimate what the computer can do when spoon fed good inputs. The problem with racing data which has always been the case is the value-added pieces aren't readily available in any consumable format, only the more generic information is readily available. The notion that I find entertaining is when people believe AI will generate its own good inputs. Like a world class chef can somehow cook great food with third rate ingredients. Bad meat is bad meat. As someone who has done my fair share of coding it's amusing what people think is possible in this arena.
|
That's the science fiction part that's evident in descriptions like this one...
https://applitools.com/blog/visual-ai/
...with hopefully metaphoric language that implies a human artifact "understanding" (rationality) and "making choices" (free will). Human beings don't even "see images in the same way", much less machines. That is why "qualia" (smell, color, pain, one's subjective visual impressions of a horse's physicality, etc.) was rejected as being objectively real in the physical world, by the Early Moderns and contemporary materialists. It couldn't be measured.
I'm not denying current or potential successes, but the potential to exaggerate or cross into sci-fi is ever present.
__________________
"I like to come here (Saratoga) every year to visit my money." ---Joe E. Lewis
|
|
|
06-11-2023, 01:10 PM
|
#14
|
Registered User
Join Date: Apr 2023
Posts: 2
|
I've been reading the boards on and off for a couple of years now and am relatively new (compared to the rest here) to betting on horses although I have bet sports for entertainment for 2 decades. In the real world I work for a large insurance company and build predictive analytic models using both "AI" and other techniques. I'll caveat this by saying that while I've given a fair amount of thought to how I think a horse race model should be built, I've not built one nor do I plan on doing it anytime soon for a variety of reasons.
The most basic question is: what are you trying to optimize with your AI model? It sounds like you are trying to employ machine learning, which can be very good at binary and classification problems. I was involved in the building a ML system several years ago that flagged medical claims for fraud. That's an example of a binary problem, the claim is either fraudulent or not.
I would suggest that ML is not the best solution (although it might play a role in stacking). A horse race involves an ordinal outcome. Horses finish in a particular order given their performance during the race and the performances of all the other horses in the race. IMO, the best modeling technique for this particular application is not ML but a simulation model.
My simulation model would have a performance probability distribution curve for each horse in the race (think of it as a speed figure or some other hybrid performance number). The simulation then bumps up against the curves to generate an outcome for each turn (simulated race). The outcome is the probability of each horse finishing in each position over thousands of turns. Along with this outcome you now have probabilities for exacta, tri, super and multi race bets.
The trick in all this of course is the tremendous amount of work needed to get the right predictive distribution curves. Just my $0.02 worth.
|
|
|
|
Posting Rules
|
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts
HTML code is Off
|
|
|
|
|