My initial observation of where I hope this all goes- my existing deT process is pretty darned good and I'm super-pleased with what it's been doing. But the final piece of the puzzle that I've been searching for is an adjudication of the low odds horses. It looks like this new Pace Model might be that final piece. So I envision this as a final step to ENHANCE but not REPLACE my existing process. — Jeff Capper

That's a reasonable expectation.

I have a question on the video pace module on the left hand side in green you have pace E pace LP and Pace SR on mine there are 8 buttons with Mdl numbered 1to 8 what does this mean and how do I change it back to the pace ones — Ira Yorn

Those are your own personal weights.


The How-to is really quite easy.
It's on the video.
Just look for the section when I'm editing the buttons.

But meanwhile...
1. You create a name.
2. Give it a button number.
3. Fill in the weights.
4. Hit Save.

Ping @Jeff Capper
Maybe when you're watching the video again, you could notice when the button discussion takes place and post it here for Ira.

1. What does pc Method 71 do differently from the important buttons that deT2 introduced? I've been relying on abPace/BH and BR2/pcrH, abRx, abRy. I've built a system that looks at all of these and gives me very good answers. I'll be re-playing recent races that I've played live to see if I can discern the differences. But any context will be interesting to know.

It uses the Model weights rather than the standard weights.
______________________________

2. I really like this new Pace Model's ability to weight each individual race. I'm now comparing these weights against the the red and green pace designation from the "front" of the AI Assistant panel. Sometimes they are consistent- meaning that the highest weighted pace button from the new Pace Model matches the green Abby Pace designation. But sometimes they don't align. Why is that? — Jeff Capper

Because it is not simply counting the ES points in the race.
Instead, it is using an AI-base algorithm.

Video is up

The only reason I'm even curious is Amwager is very professional shop, lot of the big gun rebate players live there, so I doubt they would put out junk. But I've been fooled a million times before....so, ya know — Chuck Jones

I don't think was saying they are junk.

They're just ratings. And everybody at amWager has them.

Guessing the ELO would work well for selecting contenders. — Jim Pommier

Actually, it would logically push you to shorter priced horses.

I have used chatgpt for analyzing PP's to breakdown anything in the race that a human eye might overlook. Does a great job. — Gillycapper

My point is that when people upload data to be analyzed, they think it is using a Neural Network when it isn't.

So, it can't use AI, rather it uses statistical methods-- using a lot of data. — Jim Pommier

Correct.

Yes, I watched them all. What I don't understand is that if the factors we've all been using for years only matter for low odds horses what do we handicap with?

Did I miss that in the video? — BobbyB

That's a really great question.
The answer lies in what we call HANDICAPPING OBJECTS.

It isn't that the factor ranks don't matter at all, they just don't matter at all. They still do.

They just don't matter as much as we think they do.

The answer is MORE FACTORS generated together to produce a NEW FACTOR that CAN separate the other horses.

So the two are in a single model?

↪Jim Pommier Yes. I will start a DMR model once the meet kicks in — Gillycapper

Are you modeling one track at a time?

looking forward to it keep me posted — Ira Yorn

The posting will always be as it was last night.

I'll be doing more of these impromptu meetings.

May I ask what goes into computing stretch run (SR) and best horse (BH)? — Steven

First, a little history.
My AI has two primary AI designs.

GA 47: The biggest is an engine that (essentially) created 200 "systems" and uses the "shape of the race" (as defined by the AI) to decide which system to use in each race. BTW, there are actually 2 different types - one for previous starters and one for FTS and foreign shippers.

Thus, there are actually 200 x 200 systems - resulting in 40,000 combinations.
("GA47" is a reference to this engine being my 47th Genetic Algorithm.)
________________________________
GA 37: This was a code-only system that I first developed about 15 years ago.
The purpose of GA37 was that it was MY PERSONAL TOOL for building power HANDICAPPING OBJECTS for HSH.
________________________________
Now, to your question...
I've continued developing GA37 to the point where it now does an excellent job of "integrating my specific goals directly into a fitness function." In other words, with relatively simple code, I can get "Core AI" to generate DIRECTED OPTIMIZATION towards a WELL-DEFINED SUBSET of data.

Specifically, I "told it" to bias all results in favor of horses that did not figure to be low odds.

HSH has, at it's core, about 3,500 factors, of which perhaps 500 have even an outside shot and having any real impact in the end result. (As the whales get better - i.e. the pools become more efficient - the core factors slip in value and the outliers' importance increases.

The end result was to generate 30 weighted factors for each column. These factor sets change depending upon the "AI's perceived shape of race.

Then, those factors are used at race time in a Monte Carlo Simulation.
________________________________
So, the final answer to your question...
May I ask what goes into computing stretch run (SR) and best horse (BH)?...

is, "I don't actually know."

Thanks. Plan to check out a few more races next week. Not sure if it matters, but I've been modeling only the 3-4 tracks that I wager on. — Jim Pommier

You only want to model the races you're actually betting.

In fact, only put in races that you fully handicapped and bet. (Paper bets count.)

.

.

You're correct.
I spoke to an attorney friend about how a class action suit actually works when there isn't a potential giant paycheck for the law practice. He said, it generally takes about $500k just to get to the trial date - and this was about 25 years ago.

Ryan Dickey is a fine man - I know him personally.

Could we employ the morning line in the handicapping for the method that you display here? Or do you still recommend against that, and instead to use an "artificial" morning line? Also, in the last video you mentioned an alternative way of being able to play the chaos horses. Would that involve playing all of the horses going off at above natural odds? — Dave Salvini

______________
Could we employ the morning line in the handicapping for the method that you display here?
I assume you mean in HSH.
But even not, yes, absolutely.

The only challenge in doing it by hand is that you have to adjust for scratches so that the odds aren't out of whack. In the next couple of weeks I will put up a spreadsheet that would make that easier.
______________

Would that involve playing all of the horses going off at above natural odds?
The best way is not currently available to you. (In the videos, I briefly mentioned "The Neal". That's not something you could reproduce because its 100% AI based.

But other than that, you simply take the longest MLs. Natural odds does not apply to ML, because the scale is different.

Shorter answer: "I think Longest Morning Line method is best ."

Took at stab at doing a little modeling on the races I handicapped yesterday using the Pace 5-Column approach. Not sure if I'm doing this right.
There were 34 races. 14 of the 34 races were won by a horse that did not qualify under the Pace 5-Column approach. However, 10 out of these 14 winners were Ranked 1 or 2 using factor tTrk - (Trainer: Today's Track).
Next, looked at a handful of races from last week. 14 winners that did not qualify under the Pace 5-Column approach. 5 out of the 14 winners were Ranked 1 or 2 using factor tTrk. Not quite as good as yesterday.
Again, not sure if I'm doing this right or if these results have any significance. — Jim Pommier

Now I have a little time to respond properly.

What you'll notice with this non-DET approach is that the L horses do not dominate.

There must be 2 models - L's & H's.

When you look at the L winners you'll get a drastically different look than when you look at the H winners.

That point might have been a little overlooked last night.

I think Saturday we'll do 2 sessions:
8:30 to 9:30 will be a shorter version of last night and then 10:30 to 12noon will be to actually work on modeling with the TWO spreadsheets.

I watched all the videos. Good stuff! Unfortunately, I couldn't make the meeting. — Dio

We'll get another going - probably Saturday.
We'll also work on the spreadsheet.

I can put the spreadsheet in your DropBox if you like.