beyond complexity ?

[denniswilliams]

Originally Posted by Jeff P

Having done it I'm pretty sure it's worthwhile to create algorithms that score running line data (positional calls, beaten lengths, gain or loss between calls, etc.), normalize the scores - and create running line factors from the normalized scores.

Then include some of your running line factors in a fundamental model - with many other factors.
-jp
.

Not sure we're on the same page. Not looking to normalize anything. These (especially large field turf races with multiple moves) ultimately play out in different general types. With each type having variations. And each variation favoring runners in certain positions. So, you would have had to identify these types, then, given the type, identify which horse(s) got the favorable trip, which didn't and those that just didn't run.

Not trivial by any stretch.

[classhandicapper]

Originally Posted by denniswilliams

Not sure we're on the same page. Not looking to normalize anything. These (especially large field turf races with multiple moves) ultimately play out in different general types. With each type having variations. And each variation favoring runners in certain positions. So, you would have had to identify these types, then, given the type, identify which horse(s) got the favorable trip, which didn't and those that just didn't run.

Not trivial by any stretch.

If you looking to refine it past the more generalized favored the front or favored the back, I don't think that's much of an issue from analyzing the chart. It would just involve looking at middle moves. For example, I look at where the horses were at the 1st and 2nd call in sprints and then finish, but I don't bother looking at whether a horse made a big move between the 1st call and 2nd call and how it finished. That could easily be added, but I'm not sure how much gain there is. My research indicates it takes a pretty serious extended move to have a significant impact. THat kind of thing would leap off the pages if my existing metrics didn't fully capture it. For me, that's part of the subjective process. I'd notice if only 1 horse made that move and it had an impact or if everyone that made that move was impacted.

Where it becomes more work is in the rules you personally define in order to identify horses that outperformed or underperformed what it looks like on paper. You have to have a set of rules for each of the moves based on research and experience. Mine started with experience and common sense and then evolved when I started testing the metrics and upgrades/downgrades against live races going forward.

Personally, I think you have to have averages.

The normal flow for each surface/distance is different (field size matters a bit also and I have adjustments for that). It's hard to tell whether to upgrade or downgrade something if you don't know what's normal/average for that surface/distance/field size.

If you are thinking of trying to automate the AMOUT of upgrade or downgrade based on this data kind of like TimeformUS does with pace figures, I think you are in very sketchy territory.

[Jeff P]

Originally Posted by denniswilliams

Not sure we're on the same page. Not looking to normalize anything. These (especially large field turf races with multiple moves) ultimately play out in different general types. With each type having variations. And each variation favoring runners in certain positions. So, you would have had to identify these types, then, given the type, identify which horse(s) got the favorable trip, which didn't and those that just didn't run.

Not trivial by any stretch.

Yeah, that's different than what I was talking about.

I was talking about creating factors from the raw positional calls and beaten lengths data and adding some of those to an existing fundamental model. (Also not trivial.)

I take it you're identifying various trip types from a combination of the positional calls array, race flow, or charts and video?

If so, after nailing down the rules for what it is you are capturing:

Consider adding a handful of True/False columns to your db or spreadsheet.

Column names might be: type1, type2, type3, etc. -- one column for each trip type or thing you are capturing.

Data values stored in the columns: 0 or 1... 0 that thing is False... and 1 that thing is True.

The value in doing this comes once you've recorded a few hundred races.

Depending on what you are actually capturing: It's likely many of your oservations will turn out to be useless.

But a select few could point out future overlays.

I realize stuff like this is subjective.

Provided you make the effort to arrive at the 0's and 1's in a consistent way:

Having 0's and 1's for stuff like this in a db comes in handy if you ever do decide to create a fundamental model.

Most stat packages are designed handle data having a mixture of rank, numeric value, and columns populated with 0's and 1's.

Regression (or other statistical) analysis on the data should tell you which of your 0 and 1 columns are significant (if any.)

Point is there really isn't a way to know without putting in the work.

Which again is not trivial.


-jp
.

[steveb]

Originally Posted by Dave Schwartz

I do not have a horse in the race named "Convince SteveB to Agree With Me Stakes."

I should have also said adjr2, rather than coefficient....wine must have started its job.

[classhandicapper]

I realize stuff like this is subjective.

Jeff,

I think what you are suggesting is similar to what I am suggesting.

It almost has to start out subjective.

Again, I started with an analysis of the thousands of charts to find averages for each distance/surface. Then I created a formula to evaluate each chart relative to average. I give each a rating of from between -2 and +2 standard deviations with zero being the average for that surface distance.

That was tested against pace figures independently. I wanted to make sure that two independent unalike measurements were saying the same thing most of the time. That ensured that my chart ratings were capturing reality.

Then there is an endless series of "IF" statements.

I started with stuff like this.

If Chart Rating > 1.49 and horse was 1st or 2nd within 2 lengths at the early calls that's a "Strong Upgrade Category A"

If Chart Rating > 1.49 and horse was 3rd within 3 lengths at the early calls that's an "Upgrade Category A"

If Chart Rating > .99 and < 1.5 and horse was 1st or 2nd within 2 lengths early that's a "Upgrade Category B".

There's a long list of Chart Ratings ranges tested and a long list of Upgrade and Downgrade IF statements.

Once you have all this data you can then go back and say

How did all my Strong Upgrades from Category A do when they ran back?

When the results were not satisfactory, I changed the "IF" statements to tighten the rules, reran everything, and tested the new conditions. I kept going until I started seeing consistently positive results from almost every upgrade/downgrade category.

When I struggled to find anything, I looked at each individual race within the category that failed badly and started adding exceptions.

For example, it's tougher to downgrade horses that WON their last start even if they had a perfect flow. Some of those winners would have won anyway. So now they are downgraded internally, but I'm more careful before I bet against a winner last out than say a horse that finished 2nd or 3rd that was downgraded.

I've been building this since late 2014.

Where it REALLY gets positive results is when you upgrade a horse out of certain flow and you are projecting the opposite flow the next time with a pace projector. I have my own because I need it for testing, but Timeform's is excellent.

So say you upgraded a closer that was coming out of race that flowed heavily to the front. That's already interesting, but if your pace projector is saying that today's race is very likely to have a pace collapse, then you are in seriously fertile grounds.

If the OP wants to test more detailed middle moves etc... he can do it the same way.

[Jeff P]

Absolutely.

I'm a huge proponent of creating your own metrics for stuff that's not in the Equibase data and never will be.

Or even stuff that's in the Equibase data - and despite all efforts at campaigning them to correct it:

Equibase knowingly continues to publish data on a daily basis that's bad.



-jp
.

[cj]

Originally Posted by denniswilliams

Say I have a race positional running line as follows (an actual 1400M race @ Sha Tin):

8 8 8 1
7 7 4 2
1 3 1 3
9 9 9 4
12 12 10 5
6 6 5 6
11 11 12 7
4 4 2 8
2 2 6 9
3 5 3 10
10 10 11 11
5 1 7 12

I can measure the 'complexity' of the race (the level of positional changes) using a formula but it won't 'reveal' that the horse that finished 3rd ran the best race given the setup. A huge race, in fact.

Obviously, I can look at the chart for the race and see this. And more broadly, that pace horses did not fare well.

But perhaps there's a way to automate this and not have to do as much work. Or maybe train a model to do the work for me.

Any ideas?

Maybe the 3rd finisher ran the best race, maybe not. Nothing is simplistic in this game. What if the pace was average and the other horses near the front were all big longshots and the closers were the favorites?

Looking at the flow is a factor, but not in isolation in my opinion.

[denniswilliams]

Originally Posted by cj

Maybe the 3rd finisher ran the best race, maybe not. Nothing is simplistic in this game. What if the pace was average and the other horses near the front were all big longshots and the closers were the favorites?

Looking at the flow is a factor, but not in isolation in my opinion.

This was an interesting race. The 3rd finisher won next out paying $17. The 9th finisher ran 3rd in the same race @ 49:1, beaten 3/4 of a length. But that race went 4-3-2-1 around the track on a day that appeared to favor front runners. Interested in how the winner runs back as everything else in the race backed up the last 200M in relation to its run.

As for the pace, without adjustments, based on the HKG pars for class and distance by split:

slow .1
fast .29
slow 1.36
fast .1

overall SLOW 1.07 on a day that was .62 slow average

[classhandicapper]

I leave the odds analysis separate from the flow metrics.

The public will adjust the odds based on the expected flow. So maybe those 8-1 or 10-1 or even 15-1 speed horses that didn't last on the pace were actually very good horses. They just went off longer because the public expected them to get outrun or burnt out by a fast pace and they did.

IMO, if you want to do this right you start by looking at the makeup of the field before the race is even run. You want to know how much speed is in the race and the quality of that speed.

Then you watch the race, especially the start, to see who was bumped, who was urged, who was hard held etc. That way you know if and why the race may or may not have developed as you expected.

Then you want to look at BOTH the pace figures and good quality flow metrics that compare the flow to what's typical for that surface/distance.

You need both because flow captures things that pace figures alone won't capture (number of horses engaged and how hard they were being used, short intense moves, track bias impacting the normal pace relationships, quality of horses impacting the normal pace relationships, changing track conditions etc.. ) Also, fractions/pace figures are even more prone to error than final time figures due to wind, various sections of the track not being the same and horses rating.

However, you also want to look at the pace figures, especially in races where it's less clear from pre race analysis and race watching what happened. They will help clarify what you saw and give you better grasp of how extreme things might have been.

If you do all that, you are going to have a pretty firm grip on reality for most races, but you will also wind up like me where you spend so much time trying to be right about certain races, you are going to have to narrow your play because there's not a lot of time for handicapping. I focus mostly on stakes quality dirt races around the country and the very best turf races. I could never do all this for an entire circuit and especially not multiple circuits.

I'd be happy to post a screen shot of my PPs where most of this info in located right in my Formulator notes.

[SmarterSig]

May be of interest
https://markatsmartersig.wordpress.c...al-sectionals/

[denniswilliams]

Originally Posted by SmarterSig

May be of interest
https://markatsmartersig.wordpress.c...al-sectionals/

Thanks

I use KNN to evaluate race complexity (normalized by number of starters).

Number of positional changes in a race is a good thing to know but it's not very helpful (except in obvious cases) when it comes to how the race was run.

Charting races with the same rating clearly shows this (as does just looking at the race positions)

Is there an algorithm that will not only show the change in terms of position but also where (when) it happened? The end result being that all races with a given rating were run the same way.

[crestridge]

I don't know if this is appropriate or not, forgive me if I'm off here. But many years ago I read this book on investing in the "market" called "Market Wizards", by Jack Schwager, which identified very serious successful market investors. One of the guys was identified as a "Zen" investor. In general, he just looked at the big picture, looking, at the numbers of the money movements, the elements envolved in the prospective company, in a general way. Just allowing his mind to see all the data flowing, and beginning to formulate a pattern of all this data. Just looking, observing. Some kinda way, his mind would teach him patterns.

My point here: go to the running lines of say many horses, observe past performances and just look at their rank, positions at the call, and find patterns of success. When they did well, what was the positions/ranks, and what were they doing when they did poorly? It's amazing, not writing down anything, just letting your mind tell what seems to evolve, (pattern wise) from just looking observing many horses, and what they did. If you want to expand this, also look at the tracks their running in, what class their running in. Just look at the big picture, and see if one can learn something from the "big picture". Just a crazy thought...then continue with your usual way of analyzing the races, but in the back of your mind, keep in mind what you learned from just observing the big picture.

[headhawg]

The late Michael Pizzolla wrote about "The Magic of Soft Focus" in Handicapping Magic. Seems like a similar concept.

[BarchCapper]

Originally Posted by headhawg

The late Michael Pizzolla wrote about "The Magic of Soft Focus" in Handicapping Magic. Seems like a similar concept.

That's exactly where my mind went when I read it. One of the challenges I find in using any kind of electronic PPs is not being able to "drink in" all of the PP's in one soft-focus type view. I should print out a few sets once in a while to see if that helps.