Information overload

[JustRalph]

Originally posted by LOU M.
Would you be able to use data mining software to uncover trainer patterns. Don't know how it works but I remember someone talking about it on another thread. Something like "Daisy".I'd also be interested if anyone has experience with data mining software.

I can do these ....... and more........

[Jeff P]

Lou M., JustRalph,

My own software currently does exactly the same thing as shown in the Equisim screenshot. Doing that can quickly eat a lot of free time. What I want to get past is having to run single queries for single trainers.

Let's say we are looking at a card for the first time at a new track. There's probably somewhere in the neighborhood of 40 plus trainers whose patterns we might have an interest in checking out. Who are they and what are their strengths and weaknesses? What I am in the process of creating is a set of reports that will provide answers at a glance for every trainer out there.

[Jeff P]

Building blocks for what?? What insight do you get from these realtionships.

Fastracehorse@DRF,

What I am really doing is using knowledge gleaned from research into the effects of isolated factors to construct composite ratings involving two or more factors. The relationship that exists between the factors used in the making of a composite number can have a significant effect on the effectiveness of the composite number being created.

I'll try to use my own CPace number to illustrate this point. My CPace number is made up of E1, E2, and Turn Time from selected pacelines, with running style, and Speed from selected recent workouts.

I'm able to measure the effectiveness of the CPace number by comparing the top CPace horse against the general population of horses as a whole. When I do this I can calculate an impact value for win percent as well as an impact value for ROI.

I'm also able to do this for each of the components (E1, E2, Turn Time, Running Style, and Workout Speed) used in the making of the CPace number. The thing that is very satisfying is seeing that the resulting CPace number has always been stronger in terms of win percentage and ROI than any of the single factors that are needed for its creation. This kind of told me that I was on the right track.

Now, I'll try to answer your question a little more directly.

By looking at the impact values for win percent and ROI of the individual factors (E1, E2, Turn Time, Running Style, and Workout Speed) I could see that each had a relationship to the other.

Example: You seldom have strong turn time without strong E2. But not all strong E2s are good indicators. A horse that habitually breaks poorly will often rush up to the field and end up with a really strong E2 before fading. In such a cases, you end up with a really strong Turn Time as well. CPace works best for me when E1 and E2 are not out of whack with each other. To properly construct something like what I was attempting to create you first have to have an idea of just what whack is. After studying the impact values of the single factors and the relationships they have on each other I was in a better position to know how to weight them when constructing the CPace rating.

[GameTheory]

Jeff is on the right track, and just what I was referring to earlier in this thread about causes vs. results. Handicapping is more about prioritizing then it is about weighting. One factor can be the key to everything in one race and irrelevant in another depending on the context. It is the ability to determine which factors are important in what context and why that will put you ahead of the pack...

[JustRalph]

Originally posted by Jeff P
Lou M., JustRalph,

My own software currently does exactly the same thing as shown in the Equisim screenshot. Doing that can quickly eat a lot of free time. What I want to get past is having to run single queries for single trainers.

Let's say we are looking at a card for the first time at a new track. There's probably somewhere in the neighborhood of 40 plus trainers whose patterns we might have an interest in checking out. Who are they and what are their strengths and weaknesses? What I am in the process of creating is a set of reports that will provide answers at a glance for every trainer out there.

I can write macros that will run multiple queries
but I don't personally use them. But that is just me. Some of the guys on the ES board are really into the database stuff.......

[sjk]

I'm not sure why I persist in commenting since I don't get much reaction.


The reason why I suggested focusing on an odds line rather than a composite rating relates to the ability to run lots of queries and to combine the results in non-linear and multi-step ways.


The macro I use to create my odds line uses about 200 queries.
It combines factors in a variety of ways. It seems to me that filtering everything through a composite rating would tie my hands to a considerable degree.

[GameTheory]

Originally posted by sjk
I'm not sure why I persist in commenting since I don't get much reaction.


The reason why I suggested focusing on an odds line rather than a composite rating relates to the ability to run lots of queries and to combine the results in non-linear and multi-step ways.


The macro I use to create my odds line uses about 200 queries.
It combines factors in a variety of ways. It seems to me that filtering everything through a composite rating would tie my hands to a considerable degree.

Isn't your assigned probability for each horse in your line then essentially a composite rating made up of those 200 queries?

On the lowest level, you've got the raw data. That is distilled into 200 factors via your queries (more or less, right?). Those 200 factors are then distilled into a single probability. The probabilities for each horse are then normalized over the race.

Something like that?

The above is basically what I do, anyway...

[sjk]

GT,

Not at all how I do things.

I can't think of more than a dozen or so factors. Numerous queries (and carefully computed tables of constants) are involved in calculating these factors as relates to each PP line for some factors and as relates to today's circumstances as relate to others.

Odds are derived from probability distributions. If everything comes from a single composite rating, I would expect that the distributions would have to be derived in a fairly straightforward way from the rating. I am sure that there is more than one way to skin a cat, but the distributions I use could not be derived in this way.

[LOU M.]

My software,HTR, does the same. Multiple queries, reports,etc.I have over 170 factors in my DB. Like Jeff said, you could spend a lifetime trying different queries looking for the right combo for the right situation. What I'm looking for is software that will take my DB factors and create its' own queries looking for positve combos and then alerting me to when these favorable situations are present in todays races.Is there such a product? Anyone?

Lou M.

[JustRalph]

Originally posted by LOU M.
My software,HTR, does the same. Multiple queries, reports,etc.I have over 170 factors in my DB. Like Jeff said, you could spend a lifetime trying different queries looking for the right combo for the right situation. What I'm looking for is software that will take my DB factors and create its' own queries looking for positve combos and then alerting me to when these favorable situations are present in todays races.Is there such a product? Anyone?Lou M.

Not that I know of..........yet.........but I agree it would be awesome to find.

[Fastracehorse]

<The relationship that exists between the factors used in the making of a composite number can have a significant effect on the effectiveness of the composite number being created.
===============================
In itself, the above doesn't make sense. If you are trying to say that certain factors or combination of factors should be weighted differently, then that is interesting. Especially so if you are attempting to do this.



<The thing that is very satisfying is seeing that the resulting CPace number has always been stronger in terms of win percentage and ROI than any of the single factors that are needed for its creation.

===========================

That shouldn't be too surprising Jeff. For eg., E1 and E2 factors without any semblance of pace or times/speed per se, is very limiting.




Your Cpace # - although it is your own, sounds like fairly basic handicapping to me. Most players are speed players (E1, E2, and, running style), like horses in contention mid-race ( turn times ) and like fast works.

I'm sure you have found st interesting in your intensive work. What is that??

fffastt

[Jeff P]

SJK,

I don't use an odds line per se. But I do very strongly believe in making play or pass decisions based on expected value.

My whole process is probably a little different than what you may have seen in the past. But it serves me well and I'll be happy to share what I do.

Instead of an odds line I use something that I call an odds divisor. This odds divisor is simply a number that I divide the size of the field by to arrive at the minimum required odds for placing a bet. Let's say that I am trying to make a play or pass decision for a play type that has a divisor associated with it of 1.5. In a six horse field I need 4-1 before placing a bet. In a ten horse field I want 6-1. And in a twelve horse field I want 8-1. I assign different odds divisors to different play types.

One of the things I have created in my own software is a Profile Marker module. Let's say that in my research I have identified a simple profitable play type (or Profile) that calls for a win bet whenever my top ranked CPace horse also happens to be ranked 1,2,or 3 in Form. Let's further say that this play type also requires an odds Divisor of 1.5. The values describing the play type (or Profile) are stored (semi permanently- I can edit them any time I choose) in a Profile Table. Each time my Profile Marker module looks at a new race it pulls the values stored in the Profile Table. Logic in the Profile Marker module then compares the values for each horse against the values pulled from the table. Any horse that qualifies as a play according to the Profile stored in the table is then "Marked" so that it becomes highly visible as a play on any screen or report that I have. I also have a betting instructions field in the profile table. Here, I store text that tells me how to bet this play type. This will just about always include my odds divisor.

Sample output from one of my reports taken from yesterday's card at Churchill looks exactly like this:

CD R3 7F D C
#4 TRADITIONAL/VELASQUEZ CORNELIO DS_WeightedFig 4.00 [999/min 3-2]
#4 TRADITIONAL/VELASQUEZ CORNELIO DS_Form 4.00 [2.5]

CD R5 8F D C
#2 BIRTHDAY SONG/VELASQUEZ CORNELIO DR_Overlaid_J 2.50 [1.5]
#3 PARIS LEGEND/D'AMICO A J(-+T) DR_RiderPlay 15.00 [1.5/.007]

CD R7 8F T A
#6 PUGET SOUND/DAY P(+-) TR_NiceHorses 1.20 [999/.02]

CD R10 6.5F D S
#1 KIPPER VILLE/ALBARADO R J(--T) DS_Overlaid_J 10.00 [1.5]


The top line for each race has the track, race number, distance, surface, and class code. Then I have one line for each qualifying play type. This has the post position, horse's name, rider's name and rating if there is one, followed by the name of the play type, morning line, and then the text inside the brackets comes from the betting instructions field in the Profile Table.

After running this report all I need do is compare my own betting instructions against the tote board. Everything else (as much as possible) has been previously done by my software. I try as hard as I can not to pay attention to anything else. After all, the name of this thread is Information Overload.

[Jeff P]

<The relationship that exists between the factors used in the making of a composite number can have a significant effect on the effectiveness of the composite number being created.

In itself, the above doesn't make sense. If you are trying to say that certain factors or combination of factors should be weighted differently, then that is interesting. Especially so if you are attempting to do this.

Fast,

I am saying that certain factors should be weighted differently.

In the research I did during the construction phase of my CPace number I found that the factors I used to build it (E1, E2, Turn Time, Running Style, and Speed from Recent Works) all had different impact values with respect to win percent and ROI.

When experimenting with different factor weights I found that weighting them in accordance with their impact values worked better than anything else.

After doing this I did further research and looked at races where it just flat didn't work. One of the things I discovered was that whenever a horse broke slowly and then rushed up to take the lead at the second call before fading this always resulted in a paceline that had strong E2 and strong Turn Time and therefore a strong CPace. Yet these turned out to be the horses that never seemed to compete in subsequent races. After I began downgrading this "Artificial Turn Time" associated with this phenomenon by lowering the E2 and Turn Time factor weights my results improved rather dramatically.

[hdcper]

Hi Jeff,

Glad to see you sharing some of your great ideas!!! Believe me everybody, I know Jeff personally from his days at Tup and his approach to the game is second to none.

By the way Jeff, I sent you an email or two in the private message section and look forward to hearing from you. Maybe we can chat soon if you have time.

His Cpace ratings give a good picture of the pace scenerio and several of your new ideas you have suggested just add strength to identifying solid profitable plays.

Look forward to hearing from you,

Bill (hdcper)

[Jeff P]

Your Cpace # - although it is your own, sounds like fairly basic handicapping to me. Most players are speed players (E1, E2, and, running style), like horses in contention mid-race ( turn times ) and like fast works.

What I was shooting for was one number that really expressed early speed. I didn't want to have to spend precious time while at the track handicapping to find it. I started out simply averaging the Bris E2 numbers and using those as a benchmark. The rest kind of snowballed from there.

It's not enough for me to know that horse A and B both have early speed. To me it makes a real difference to know which one is faster and by how much. The first question I try to answer when looking at a new race is "Who gets the lead in here?" When watching races being run, I often see lesser ranked CPace horses get in front of the top ranked horse, but they almost always expend too much energy to do so.


I'm sure you have found st interesting in your intensive work. What is that??

What I found really interesting was the number of longshot winners I was getting using early speed compared to anything else. From 1996 when I started using it to as recently as 2001 it seemed like $100.00 early speed winners were not all that uncommon. In recent years I haven't had one. I think this can be attributed to the growing number of players using Bris and other pace ratings in their handicapping. The game changes constantly. Early speed doesn't pay as much as it used to. But believe me, it still pays pretty well.

Another thing I find very interesting is how this one number helps me identify track biases. At some tracks, for days at a time, it seems like the only horses able to win are those highly rated in CPace. At other tracks, and at other times, the opposite is quite true. I think this can be very crucial information to have when making a play or pass decisions on a horse.