Programming intuition, creativity and speculation

[dnlgfnk]

Once upon a post, "Just Ralph" essentially stated to me that any handicapping idea could be programmed. That concept occasionally returns to me, as it did after the 2nd race at Santa Anita today, wherein I had a successful interpretation of the race.

I've no concern or care for suspicions of "redboarding". The issue for me is the capacity for programming the aspects of thought mentioned in the title. In preparing for the coming handicapping year, I would love to be able to arrive at conclusions with more speed and quantity.

The following is my potential trip scenario, followed by superficial comments which serve as a type of "search engine", whereby I arrive at deeper subjective conclusions. They have been formed over time and experience, i.e., a mental database compiled largely by years of observing horses' finishing positions relative to their odds. I assign confidence values to the various significant aspects of the comments/factors (as well as tote-smoothing), or more to the point, the deeper conclusions they lead me to.

In the hunt for overlays, I begin with a negative perspective (red) of every horse, beginning with the ML favorite. Inversely, I establish any positives (light green) for the field beginning with the longest odds, matching those factors against the still negative view of the more probable horses in order to justify the positives of longshots, if able.

7
2 456
3
1

R2--SA 8F 4+ A40k(N2L)

5 Miss Star Maker) broke from outermost pp last 2 races; easy win vs m20k field (79 win fig);draws outer speed of #7...potential duel or outrun for lead by that sprinter/ speed threat has stalked at 7f; pp5 of 5 last out--draws inside main speed foe; favorite.

4 Blondy's Reward) lacks high speed but races only evenly--draws inside #'s 5-7 to ft/ competitive figs; draws inside big speed, force their effort to ft; stalk for piece of exotic?

2 Bijoux) speed vs M20k @ Lrc (79 win fig); draws much speed from outside--closed 4l 3 races ago/ has closed 4L previous @ 1m--could potentially stalk, save ground turns; force pace to ft inside 4 other speed foes.

3 My Sweet Afleet) not close last 2 races; last 8 races = 0 for 8 vs A40k/nw2l...last 5 hasn't shown early speed; draws pp3 of 7 after last closest recent fin was wide rally/ gets contentious pace; strong closing ability vs winners; #7 takes out #5 & #4 regarding those common opponents in last 2.

7 Emotional Drive) sprinted to win at 5f vs M20k @ Lrc (81 field); stops at 5.5f when not clear lead; outside four other potential speed rivals to ft/ superior speed; utilize outside footing to ft; burn up other speed.

6 Nazareth) sprinting last 7 races; draws sprint speed of #7 & inside her; seems to need lead w/o able to get it in sprints- even at 7f; 0 for last 7 vs $25kclm(nw2L); Chantal?/high stalking potential clear 2nd possibly inside--secondary sprint speed to #7--can keep her outside or force effort to ft.

1 Lindy Hoppin) lacks early speed vs M20-M26k @ GG; draws pp1 even more slowish brk; poor pct. jck-trnr; best fig 78 & still mdn; closest fin 3 by 3/4 to M20k (73 field)/ gets very contentious pace or lone leader; potentially saves on turns.

My Sweet Afleet (#3) was a strong overlay based largely upon the potential pace, her capability of closing vs. equal or better fields, and other minor factors. The 2nd finishing favorite was designated below the winner due to the virtual impossibility of reproducing the win of two races ago with the presence of #7, plus the tendency to seek the front at 1m despite stalking in sprints. She scored strongly most by tote smoothing.

Anyway, with my subjective factors constantly changing their value or even applicability from race to race, and the manner in which language such as I have written in the comments section fails to represent the depth of the actual thought process involved (intuition, creativity, speculation, etc.), how can that endeavor be possibly managed into a technological endeavor?

[Dave Schwartz]

how can that endeavor be possibly managed into a technological endeavor?

You have raised the bar to a very high level.

Not only must the data be interpreted properly with some very significant A.I., but the first thing you must have is a completely separate layer of A.I. just to understand the words.

Natural Language Processing (NLP) is a field of A.I. all unto itself.

The project you are talking about is being developed at places that receive multi-million dollar grants.

[dnlgfnk]

Originally Posted by Dave Schwartz

You have raised the bar to a very high level.

Not only must the data be interpreted properly with some very significant A.I., but the first thing you must have is a completely separate layer of A.I. just to understand the words.

Natural Language Processing (NLP) is a field of A.I. all unto itself.

The project you are talking about is being developed at places that receive multi-million dollar grants.

In other words Dave, in our lifetimes, the deeper that one (not necessarily me) possesses the capacity for the type of insightful thought I described, the less his sense of the race can be represented technologically?

I wouldn't be content knowing that a significant factor was in play for this single-race event, that I couldn't convey to a programmer due to the limits of speech vis a vis our deepest sense of experiential/intuitive knowledge.

[thaskalos]

Originally Posted by dnlgfnk

In other words Dave, in our lifetimes, the deeper that one (not necessarily me) possesses the capacity for the type of insightful thought I described, the less his sense of the race can be represented technologically?

I wouldn't be content knowing that a significant factor was in play for this single-race event, that I couldn't convey to a programmer due to the limits of speech vis a vis our deepest sense of experiential/intuitive knowledge.

Welcome to the ranks of the discontented.

[Hoofless_Wonder]

I don't see why spoken language has to be turned into code via AI. For the above example, much of the logic can be programmed into calculating the pace for race, positions of each horse in the projected pace, and then closing positions and late pace. Combine that with some variables assigned to each horse on how their past races will indicate success with the pace scenario, and it's not difficult to see the go to the lead and fade in a faster than average pace, with the taking command at the quarter pole and drawing off.

Now I'm not saying this is easy, as it takes plenty of code just to come up with a decent pace model, and of course many more lines of code to model in the "angles" and analysis of each of the horses' PPs. You'll have to live within the limits of some races not being very predictable, or the occasional factor that causes the race to unfold in a completely different way (i.e., jock falls off the lone speed horse coming out of the gate). If the main goal is to overcome the communication gap between your handicapping approach and a programmer's ability to convert that to a usable algorithm, then again, it won't be easy. Even when programming it yourself.

But after a while, all of the common outcomes are covered via code, which takes into account much of the experience, intuition, creativity and speculation related to predicting the finish. The difficult part then becomes weighting the inputs (which of course can change from race to race), and tracking each "intuitive" portion of the code to determine its impact and reliability. Depending on the success and accuracy of the code, it should become evident on which areas need to be improved.

Based on what I see in the above example, there's not any aspect of that, as far as I can tell, that couldn't be programmed.

[Dave Schwartz]

In other words Dave, in our lifetimes, the deeper that one (not necessarily me) possesses the capacity for the type of insightful thought I described, the less his sense of the race can be represented technologically?

Natural language is where the bar leaps.

Thus, if you had an observer trained to quantify thoughts into data, you could make the entire process about 2 magnitudes easier.

I don't see why spoken language has to be turned into code via AI. For the above example, much of the logic can be programmed into calculating the pace for race, positions of each horse in the projected pace, and then closing positions and late pace. Combine that with some variables assigned to each horse on how their past races will indicate success with the pace scenario, and it's not difficult to see the go to the lead and fade in a faster than average pace, with the taking command at the quarter pole and drawing off.

Easy answer... if it is easy, why has it not been done?

[BCOURTNEY]

Originally Posted by Dave Schwartz

Easy answer... if it is easy, why has it not been done?

The technology in question is NLG - Natural Language Generation.
It's a way of story telling from data. These stories can then be fed into an NLP system of analysis... or presented to humans in any event.

https://www.narrativescience.com/
https://www.narrativescience.com/sports
https://gc.com/

http://www.arria.com/

Some people already have technology generating alternative text charts of horse races strictly from the data, chart writers have consistency issues. Some of these alternative computer generated charts can even turn into a computerized voice audio and visual (a virtual recreation) of the race in question - Happening already.

[BCOURTNEY]

https://www.recordedfuture.com/recor...e-day-trading/

[BCOURTNEY]

Would anyone pay for a web based service to enter a date, track code and race number to receive a text chart of the race as interpreted by machine?

What if the computer could tailor the responses in a way the recipient wanted?

More importantly what if the race text was viewed prior to a human watching a video replay or crowd-sourced to many people to ensure its correctness and assist in teaching it an a semi-supervised manner?

Would be great fun to see a word cloud for every horses past lines prior to a race.

There are a few using this technology to conduct day trading strictly off text only posting surprisingly positive returns for stocks.

[Dave Schwartz]

Those are interesting links. Thanks.


(Truthfully, I don't care much about the language. I am more interested in applying the principle to conventional factors.)


I will be mulling.

[Hoofless_Wonder]

Originally Posted by Dave Schwartz

Natural language is where the bar leaps.

Thus, if you had an observer trained to quantify thoughts into data, you could make the entire process about 2 magnitudes easier.



Easy answer... if it is easy, why has it not been done?

IBM created Watson - it's pretty good at chess Jeopardy, but it wasn't easy getting there.

Based on the example given, a programmer with some racing background, or a handicapper with some basic programming knowledge could bridge the gap necessary to translate thought processes into appropriate algorithms. Compared to creating a "Watson" for the track, this is easy. I didn't mean to imply that NLP was easy.

[dnlgfnk]

I've glimpsed examples of this in recaps of fantasy football games provided to the combatants. The recaps read like articles in any major newspaper's sports section, including sarcasm and humor.

In the race I cited, my subjective analysis identified three potential overlays...#'s 1, 3 and 6. While #2 Bijou had closed those 4 lengths at a route on 10/15, she surrendered position to two rivals thru the stretch and was highly likely to participate in the pace before the stretch turn. As mentioned, #3 was solid, but I wrestled with #6 Nazareth.

How would this AI language, or a decision tree?, mimic my thought process which was fluid, visual, intuitive and speculative, rather than data derived or analytical, as I wrestled with whether Chantal would press hard to keep #7 Emotional Drive outside and under urging to the first turn, or else allow others inside her to do the dirty work while stalking to the far turn, as my prompting notes of Sunday steered my thinking ("high stalking potential clear 2nd possibly inside--secondary sprint speed to #7--can keep her outside or force effort to ft.")-- in this case to decide upon her inclusion and emphasis in the exotics. And how could it do so in countless other trip scenario situations that arise?

[thaskalos]

Originally Posted by dnlgfnk

Anyway, with my subjective factors constantly changing their value or even applicability from race to race, and the manner in which language such as I have written in the comments section fails to represent the depth of the actual thought process involved (intuition, creativity, speculation, etc.), how can that endeavor be possibly managed into a technological endeavor?

When JustRalph told you that "any handicapping idea could be programmed"...I doubt that he was talking about the handicapping ideas which are "constantly changing their value, or even applicability, from race to race".

[Dave Schwartz]

How would this AI language, or a decision tree?, mimic my thought process which was fluid, visual, intuitive and speculative, rather than data derived or analytical, as I wrestled with whether Chantal would press hard to keep #7 Emotional Drive outside and under urging to the first turn, or else allow others inside her to do the dirty work while stalking to the far turn, as my prompting notes of Sunday steered my thinking ("high stalking potential clear 2nd possibly inside--secondary sprint speed to #7--can keep her outside or force effort to ft.")-- in this case to decide upon her inclusion and emphasis in the exotics. And how could it do so in countless other trip scenario situations that arise?

It would not do that.

It would create its own "thought process."

The only way it would mimic yours would be if you fed it questions where the goal was not the "right" answer but rather "your answer."

The problem is that I doubt that you could actually have a single answer.

As an example, ask yourself what is your ultimate answer for a given horse? Is it something like, "On a scale of 1-5, I give this horse a 5?" That is a good example of quantification that software could possibly come to understand.

But that isn't what most people do. Usually there is no clear output, something that a software program needs - even if it is a bit "fuzzy."

Take the "1-5" question. Software could actually deal with it is 2-ish. That could be fuzzified with a little effort. But the human must make the effort to do that.

[dnlgfnk]

Originally Posted by Dave Schwartz

It would not do that.

It would create its own "thought process."

The only way it would mimic yours would be if you fed it questions where the goal was not the "right" answer but rather "your answer."

The problem is that I doubt that you could actually have a single answer.

As an example, ask yourself what is your ultimate answer for a given horse? Is it something like, "On a scale of 1-5, I give this horse a 5?" That is a good example of quantification that software could possibly come to understand.

But that isn't what most people do. Usually there is no clear output, something that a software program needs - even if it is a bit "fuzzy."

Take the "1-5" question. Software could actually deal with it is 2-ish. That could be fuzzified with a little effort. But the human must make the effort to do that.

Dave, thanks for your input.

You've helped me realize that it is me trying to emulate the software. I am in the early stages of scoring every piece of information I deem significant in terms of a "confidence value". And I'm very interested in the "2-ish" area, remembering that Benter effectively asked if the distance factor is solved by yes/no, 0 or 1?

Yes, it is effort and more selective action. But I hope to emphasize the units I wagered in the past.