Data Shaping Problem

[cj]

Originally Posted by thaskalos

If the horse appears equally proficient on turf and dirt...would you consider using a recent dirt race in your handicapping of today's TURF race, and vice versa? I sometimes run into such a dilemma when assessing current form.

I do this all the time. Of course it depends on price, but just had one on Belmont that had the best last race figure in the field. It was from a turf race, today was on dirt. So was the improvement a move to turf, or improving form?

I scroll down and find her other turf start and see her fig was similar to dirt races before it, so why not give her a shot? She had other things too, being a closer in a race we had as likely to have a fast pace. But like most things, if she is 7-2 It would temper my enthusiasm a lot. But 10-1? I'm in. I'll lose that bet 17 or 18 times out of 20, but I'm ok with that.

[classhandicapper]

Originally Posted by Dave Schwartz

Why stop there?
Age, surface, distance, pace pressure, class level.

There is already too much segmentation. If you run through all of those as segments, you wind up building several thousand systems. (Not exaggerating.)

What I am building is an AI solution that handles all of this with one system.

IOW, it is (theoretically) capable of "understanding" the nuances and putting together the right components to handicap every single race.

THEORETICALLY really is a big word.
We shall see.

I agree that you can take it too far, but "day of the week" strikes me as more related to the type of races carded on that day than anything else. High quality races are way more likely to be carded on weekends for example. However there will be times when that is not true. So you may as well try to get at the core reason instead of using day of the week as a substitute.

[classhandicapper]

Originally Posted by thaskalos

If the horse appears equally proficient on turf and dirt...would you consider using a recent dirt race in your handicapping of today's TURF race, and vice versa? I sometimes run into such a dilemma when assessing current form.

I think the longer the horse has been running on a different surface, the less likely he is to revert back to the form he showed previously on the other surface even if he's in form now. I also don't trust speed figure comparisons as much on surface switches. The pace and other differences make those comparisons iffy. I think you need to be compensated extra in cases like that.

I just had a conversation with a trainer about this. He said he's had horses that worked great on dirt, ran fine on dirt, he switched to turf because he thought the horse might like it better, the horse moved up on turf, and now works and runs poorly on dirt.

[ReplayRandall]

Originally Posted by classhandicapper

I agree that you can take it too far, but "day of the week" strikes me as more related to the type of races carded on that day than anything else. High quality races are way more likely to be carded on weekends for example. However there will be times when that is not true. So you may as well try to get at the core reason instead of using day of the week as a substitute.

I had a solid convo in a different thread with Mark from MTN about knowing the track maintenance schedule and changes to the surfaces, maybe overhearing the Supv. giving instructions to his crew or catching a memo on the track website....IMO, this is a "key" that is overlooked by the vast majority of 'cappers, nationwide.

Routine and day of the week of when certain aspects of track maintenance are performed, might land you on a "trend is your friend" discovery. Different tracks have different work patterns and routines. No one size fits all here, must do the digging.....Just my couple of cents.

[cj]

Here is another that won turf to dirt at a big price. Was she just improving, or was it turf that made her run a top number last time?

I didn't use this one at all because she looked too slow even if you assume both surfaces were fine for her. It is a tricky thing these surface switches, but improving 3yos are always worth a look at a big price and I can see why a few people I know liked her.

[ranchwest]

Originally Posted by classhandicapper

I agree that you can take it too far, but "day of the week" strikes me as more related to the type of races carded on that day than anything else. High quality races are way more likely to be carded on weekends for example. However there will be times when that is not true. So you may as well try to get at the core reason instead of using day of the week as a substitute.

There's only two possibilities I can see. The Racing Secretary writes different races for different DOW. There is a track maintenance schedule. The horses don't know what DOW it is... inherently, intellectually, horoscope wise or any other way. I don't anticipate tracking DOW for any reason because it is fraught with breakdown possibilities.

Reminds me of the track handicapper giving a seminar and saying the horse had a wide trip and trigonometry tells him the horse will run better with a better trip. My response is that horses don't know trig. They have patterns that get them into the same race situations over and over.

[geroge.burns99]

Originally Posted by ranchwest

There's only two possibilities I can see. The Racing Secretary writes different races for different DOW. There is a track maintenance schedule. The horses don't know what DOW it is... inherently, intellectually, horoscope wise or any other way. I don't anticipate tracking DOW for any reason because it is fraught with breakdown possibilities.

Reminds me of the track handicapper giving a seminar and saying the horse had a wide trip and trigonometry tells him the horse will run better with a better trip. My response is that horses don't know trig. They have patterns that get them into the same race situations over and over.

Agree....

The only thing that is different is CLASS of races....

Stat wise you have a better chance to beat the ML favorite on Mon-FRI than on weekends...( 40% to 37% ) because of the type of races that are running .

GB

[Track Collector]

Another DOW consideration.

With a higher portion of less-sophisticated monies in the pools on Fri/Sat/Sun (due to the participation of more causal/recreational players), I suspect the Return on Investment for higher-skilled players is higher those days.

[classhandicapper]

Originally Posted by Track Collector

Another DOW consideration.

With a higher portion of less-sophisticated monies in the pools on Fri/Sat/Sun (due to the participation of more causal/recreational players), I suspect the Return on Investment for higher-skilled players is higher those days.

That's probably true.

I always felt I get better value on some of my personal notes at Saratoga than AQU or BEL because there's more casual fan money in the pools. The same is true on some other really big race days that draw out casual fans.

[Jeff P]

Originally Posted by Track Collector

Another DOW consideration.

With a higher portion of less-sophisticated monies in the pools on Fri/Sat/Sun (due to the participation of more causal/recreational players), I suspect the Return on Investment for higher-skilled players is higher those days.

Agree with this. Especially on Saturdays.


-jp

.

[ranchwest]

Originally Posted by Track Collector

Another DOW consideration.

With a higher portion of less-sophisticated monies in the pools on Fri/Sat/Sun (due to the participation of more causal/recreational players), I suspect the Return on Investment for higher-skilled players is higher those days.

Very true, though that is more of a wagering factor than a handicapping factor.

[Dave Schwartz]

Originally Posted by Track Collector

Another DOW consideration.

With a higher portion of less-sophisticated monies in the pools on Fri/Sat/Sun (due to the participation of more causal/recreational players), I suspect the Return on Investment for higher-skilled players is higher those days.

Thank you, Track Collector, for making my EXACT point!

Namely, that the handicapping is DIFFERENT on weekends and, as such, it is not simply about higher class races.

An even more extreme example would be the BIG RACING DAYS. Specifically, the Triple Crown and BC days.

And not just at the host tracks!

I would imagine that the crowd at (say) CBY on one of those big race days is DIFFERENT than a typical Saturday at CBY.

[classhandicapper]

Originally Posted by Dave Schwartz

Thank you, Track Collector, for making my EXACT point!

Namely, that the handicapping is DIFFERENT on weekends and, as such, it is not simply about higher class races.

An even more extreme example would be the BIG RACING DAYS. Specifically, the Triple Crown and BC days.

And not just at the host tracks!

I would imagine that the crowd at (say) CBY on one of those big race days is DIFFERENT than a typical Saturday at CBY.

I don't think the "handicapping" is different beyond the fact that the races are of different quality. It's the opportunity to find overlays that might be a bit different. I guess if you are using the live odds as part of your line making in your system you may have to make the slightest of tweaks to the formula on different days, but a Grade 1 on a random Tuesday is the same as a Grade 1 on Saturday Breeder's Cup day other than possibly the odds.

[Red Knave]

Originally Posted by Dave Schwartz

My specific question is:
"Does anyone have an idea how to weight the training data to address all 3 of these?"

I think the time decay element is the most difficult. I also think classhandicapper has the best answer, trial and error.
You must already have a neural net component to your AI and also lots of data sets so testing various values to see outcomes seems like a reasonable starting point. You also have things like Fibonacci, Natural Logarithms, Exponential Decay etc. to generate sort-of logical values to seed the trials with.
I know you're having fun.

[Dave Schwartz]

Originally Posted by Red Knave

I think the time decay element is the most difficult. I also think classhandicapper has the best answer, trial and error.
You must already have a neural net component to your AI and also lots of data sets so testing various values to see outcomes seems like a reasonable starting point. You also have things like Fibonacci, Natural Logarithms, Exponential Decay etc. to generate sort-of logical values to seed the trials with.
I know you're having fun.

Indeed I am. <G>
Most I have had in like 10 years.

And, yes... Fibo, Logs, ExpoDecay are all parts of it.

But, in the end, I got a great piece of advice that came up in a session with one of my power users. Simply put, he asked a really powerful question:

Do you have a mechanism to eliminate some races from the training database?

I think THIS will be far more important than my original question.

The logic to remove some races is obvious - like a DQ in one of the top 2 positions or a low-odds horse who never makes the 1st call (i.e. broke down).

But I can do better than that.

The rationale behind this is anything that could not be foreseen in advance is highly questionable and the lower the odds on the horse(s) the greater the argument for excluding the race.