Question for software creators/customizers

[markgoldie]

If you are generating some sort of overall strength number ala Bris Prime Power, how do you handle layoffs? And, for example, is there a material difference between a 100-day layoff, say and 200 or 300 days? Also, do you factor in positive adjustments for 2nd off a layoff, third off a layoff, etc.?

The crux of the question is do you make "one size fits all"? Or, for example, do you also factor in age/sex/running style to the adjustment?

Thanks.

Mark

[CBedo]

For me, as you would expect probably, the answer is it depends. Sometimes a layoff is a positive, sometimes it's a negative. For different circuits, tracks, and classes of horses, & even trainers, different length layoffs mean different things. Even though we have year round racing now, there are still even times of the year where a layoff might mean something different.

As far as is there a difference between 100, 200, & 300 days, for me, they tend to have similar impacts (95% of the time roughly), especially 200 and 300. 100 can be a bit different sometimes.

I haven't done as much work with regards to running styles and layoffs, and only slightly more with sex, but absolutely different age groups of horses are affected differently by layoff times.

I guess that's the long rambling version of "there is NO one size fits all"--unfortunately!

What do you think Mark?

[markgoldie]

For me, as you would expect probably, the answer is it depends. Sometimes a layoff is a positive, sometimes it's a negative. For different circuits, tracks, and classes of horses, & even trainers, different length layoffs mean different things. Even though we have year round racing now, there are still even times of the year where a layoff might mean something different.

As far as is there a difference between 100, 200, & 300 days, for me, they tend to have similar impacts (95% of the time roughly), especially 200 and 300. 100 can be a bit different sometimes.

I haven't done as much work with regards to running styles and layoffs, and only slightly more with sex, but absolutely different age groups of horses are affected differently by layoff times.

I guess that's the long rambling version of "there is NO one size fits all"--unfortunately!

What do you think Mark?
Well, I guess I could have posed the question in a better way. What I was driving at is that guys who use a total number for a horse in handicapping a race, have to put something into the program that will handle all situations. That's because we need one number for each horse.

Now. It's all well and good to say that every situation is different and has to be handled differently. But suppose we're looking to crunch large numbers of races to come up with conclusions. We don't have the luxury of sitting there and spending a half-hour mulling over the pp's for a given event.

So, let's take the Bris Prime Power number as an example. I'm wondering how sophisticated they are with it. Obviously they must use some sort of formula so that they can crank out the final number. As a close watcher of this number, I can tell you that there is no question in the world but that they are downgrading horses due to layoffs.

Now, if they are using a more sophisticated formula regarding missed time, I'm wondering not only what it is, but the research involved in coming up with the formula.

From the top of my head, using my experience as a guide, I'd say that amount of time off, age, sex, and running style all have a bearing on an absentee's prospects when the horse returns. Also, the trainer record with layoff horses is important as well.

From what I have seen anecdotally, the younger horses fare much better than the older ones after a significant layoff. In fact, I like returning 3 year olds particularly when their last races were earlier in their 2 year old career. Seems that the time to mature can really help. On the other hand, a returning 8 year old is going to need some racing to get back into top condition. I think the speed horses fare a bit better than the dead come-from-behinders since the latter seem to need their stamina to race their best while a speedball may just outfoot a field and put them to sleep even though tiring late. The females seem to fare better than the males in their first back attempts. Possibly they need less conditioning or maybe they respond better to potentially greater soundness (which may come after a rest).

Anyway, that was what I was driving at. I could have put it another way and asked if anyone had done extensive research into this.

Mark

[Handiman]

The question isn't so much about the length of the layoff, but how has the horse been readied to comeback and run.

handi

[CBedo]

Well, I guess I could have posed the question in a better way. What I was driving at is that guys who use a total number for a horse in handicapping a race, have to put something into the program that will handle all situations. That's because we need one number for each horse.

Now. It's all well and good to say that every situation is different and has to be handled differently. But suppose we're looking to crunch large numbers of races to come up with conclusions. We don't have the luxury of sitting there and spending a half-hour mulling over the pp's for a given event.I guess I should have structured my answer differently as well. My point is that I do try to get down to one "power" number--it's the probability of winning (fair odds line). In doing that, I do everything I said before. To simplify it somewhat, think about it in a decision tree type format to get to the single rule. The complexity of it comes from their being lots of nodes in the tree, so to speak. For example, a 2 year old high class maiden layed off into his three year old season will follow a different decision path than a 5 year old low level claimer midyear who has been laid off the same amount of time (actually there would be at least four nodes there: the age of the horse, the class, the length of the layoff, and date of the layoff).

I hope that explains it a little better. And as far as research, I try to think of something logically, and then try to do the research to confirm or disprove (happens ALOT) the hypothesis. Also, around certain projects, I'll do some data mining to look for factor relationships as well. The size and the quality of the sample is always an issue.

[thoroughbred]

If you are generating some sort of overall strength number ala Bris Prime Power, how do you handle layoffs? And, for example, is there a material difference between a 100-day layoff, say and 200 or 300 days? Also, do you factor in positive adjustments for 2nd off a layoff, third off a layoff, etc.?

The crux of the question is do you make "one size fits all"? Or, for example, do you also factor in age/sex/running style to the adjustment?

Thanks.

Mark
Good Question.
To address this, CompuTrak provides a workout Form Rating based on recent workouts to, among other things, assess a horse's form (Readiness) after a layoff.

[markgoldie]

I guess I should have structured my answer differently as well. My point is that I do try to get down to one "power" number--it's the probability of winning (fair odds line). In doing that, I do everything I said before. To simplify it somewhat, think about it in a decision tree type format to get to the single rule. The complexity of it comes from their being lots of nodes in the tree, so to speak. For example, a 2 year old high class maiden layed off into his three year old season will follow a different decision path than a 5 year old low level claimer midyear who has been laid off the same amount of time (actually there would be at least four nodes there: the age of the horse, the class, the length of the layoff, and date of the layoff).

I hope that explains it a little better. And as far as research, I try to think of something logically, and then try to do the research to confirm or disprove (happens ALOT) the hypothesis. Also, around certain projects, I'll do some data mining to look for factor relationships as well. The size and the quality of the sample is always an issue.
I should say that I am the farthest thing from a computer programmer that you can possibly imagine. But I am fascinated with what you guys do, so I ask a lot of questions.

Your answer about the decision tree with multiple branches makes a lot of sense. We want to move or process the horse through the different circumstances involving his return to the races. But I guess I still don't quite know how we decide the amount to weigh each branch the horse goes through unless we have done some extensive research on the question.

You did mention data mining for some projects and I would imagine this is how the "weighing" is determined. I suppose you would look at large numbers of returning horses, put them into categories and then run either ROI numbers or maybe impact-value numbers on them. Once you find out how favorable or unfavorable a given category is, you could then "score" the horse depending on the number of categories he fit. Of course, this would only be an adjunct to evaluation of the lines you do have. But the reason I think this is so critical is that the lines we're looking at have much less importance because of the time missed. In other words, they are ancient history (in many cases) when it comes to determining the animal's current form. And having said that, I guess the trainer record with layoffs and the workout pattern are also very important, as is any past layoffs the horse has undergone (and how he fared under those circumstances). All of these factors should be included as tree branches, if possible.

At any rate, I'm just rambling on here.

Mark

[cj]

This may seem backwards, but I've found horses that layoff while in top form are poor bets when they come back. Why lay a horse off when going well?

Horses that tailed off and take a break needed the freshening. Many times they come back and run back to their best.

[CBedo]

Mark, you touched on a number of important (to me anyway) things in your last post, and I want to discuss it in more detail, but right now, between a few cocktails and my girlfriend yelling at me for the tv remote, I wouldn't do it justice. I promise I'll respond tomorrow.

I'll just say for now that you definitely hit the nail on the head when thinking about figuring out today's current form. History is only useful in so much as a) we learn from it so we don't make the same mistakes & b) using it to help predict the future. To me, that's true in anything, not just horse racing. OK -- that sounded deep when I was typing it, now it sounds corny, must be the Grey Goose kicking in.

Until tomorrow,

[Dave Schwartz]

Chris,

Gads, man! Never give them the remote! You're better off just going into the other room and changing the channel for her! Women are just not equipped for handling remotes.


Regards,
Dave Schwartz

[asH]

Chris,

Gads, man! Never give them the remote! You're better off just going into the other room and changing the channel for her! Women are just not equipped for handling remotes.


Regards,
Dave Schwartz

LOL

layoffs are strictly dependent on how connection's viewed the potential of the horse before he went to bed, then trainer prep stats.
asH

[CBedo]

This is going to be long and rambling. Be Warned! Here's a the summary:

-What is the point of software devleopment, and how should it be judged
-What research (and experience) underlies the development
-Logical design versus black box data mining
-Relative weighting of factors
-The public knows more than my model.


First off, the standard I judge everything by is my non-automated handicapping performance. Give me the form, pen, spreadsheet, software tools, and whatever other information I can get, and really dig into a race. This is the baseline. I consider any software I develop (or buy) as a tool to do one of two things: 1) close the gap between my long term non-automated edge, and 2) make the process faster, thus increasing the number of plays relative to what I could do if I did it all by hand. Don't donwnplay the second part. For example, if my personal handicapping could generate a 100% roi, but I could only handicap one track per day, and only got a couple key plays, there's a good chance that a software program that only generated something that was 10% as profitable, would more than likely generate significantly more TOTAL dollar profits if it could look at every track every day and give me automated plays. The middle ground is a software tool that would generate more potential plays than you could by hand, and then in the same amount of time you would spend on one track, you get to look at 20 or more possible plays to filter down.

In thinking about research and factors, I think it's smart to again start with the "handicapping by hand," and ask the same questions (as many should). My methodologies have come from what I was originally taught, what I learned, what I've researched, what I've found doesn't work and tossed, etc. Is my handicapping as good as possible--no chance. In fact, interestingly, the more research I do on software development to close the gap between my base roi and the software's, the more I learn which hopefully raises the bar.

The research I do tends to stem from logic. For example (on the topic of layoffs), I'm sure most of us at some point either read or were told "throw out a horse that has been laid off 30 days." So you start with that basis until you figure out that it doesn't work--or it works "sort of" in certain circumstances. I take these types of hypotheses and try to confirm or disprove them. Then usually what happens is that in the process of doing that work, the results give you five more questions and areas of research. Data sample quality and size is always an issue, and whenever I can't figure something out, I try to blame it on "not having enough data!" :rolleyes: haha. As data acquisition costs fall, I'll have to find another excuse.

I have done some "throw a bunch of sh@# against the wall" data mining to see what sticks, but haven't had a huge amount of success (yet?) for the most part doing it that way. What data mining has done instead of finding new relationships, is help me understand relationships and factor weights better, especially relationships which are nonlinear (and my simplistic brain has a hard time with). I have just scratched the surface here, and continue to do more in this area to spur more research (this is research not on how to play, but research on possible new ways to develop ways to play--research on research techniques, if that makes any sense).

With regards to weighting, here's what I've learned (and as I learn more, this viewpoint could evolve or change). I'm not smart enough to absolutely weigh how important having a factor is (like a 30 day layoff), and I'm definitely not smart enough to weigh how important a gap in a continuous factor is (like 2 point speed point advantage). What I have figure out though is relative weighitng--ie, a horse with a 2 point speed point advantage (given everything else equal), wins more than the horse that is two points shy of the best--duh!. This might seem obvious, but it underlies how my weighting schemes tend to work. Let's say we have a two horse race (it's in California, hehe). Before I start the process they both are 50% to win, but as I work through different factors and relationships, each gets upgraded or downgraded, and eventually, you end up with a probability (I then will try to test and research the overall methodologies to see if the probabilities are anywhere close to reality (most times they're not), and iterate again (as part of this I have begun to get a better understanding of "fitness," not just edge, edge/odds or impact values).

The other thing related to even the best models (I think) is that the public odds will always reflect some things you didn't know (and maybe some misinformation). The Wisdom of Crowds is a real factor, and I have found that the horses I make 2/1 do win more than the ones I make 3/1, etc, BUT if you look at a large sample of horses I made 2/1, the ones the public make even money do win more than the ones the public makes 3/1 (hopefully not linearly), so I use (sometimes) a form of Bayesian inference to take into account "new" public information (Dave has talked about this in the past I think).

Now that I read over all this, it is clear that I don't know very much, and that I still have alot of work to do, and it is now more clear how truly tough all this is, to implement, and to explain. :bang: Anyone who was looking for a reason to put my posts on the ignore list--this might be it! I know this still doesn't probably answer some of your questions, but if you keep asking, I'll keep trying...as long as you return the favor so we all learn to win and win more! I love the intellectual challenge of betting horses, strike that; not betting horses, it's the intellectual challenge of making money betting horses!:ThmbUp::ThmbUp:

[Dick Schmidt]

Ah, Dave. You won't mind if I forward this to Beth, will you? Been nice knowing you.

Your Pal,

Dick

Corduroy pillows are making headlines.




Chris,

Gads, man! Never give them the remote! You're better off just going into the other room and changing the channel for her! Women are just not equipped for handling remotes.


Regards,
Dave Schwartz

[CBedo]

Chris,

Gads, man! Never give them the remote! You're better off just going into the other room and changing the channel for her! Women are just not equipped for handling remotes.


Regards,
Dave SchwartzGood call on your part! I have a Logitech Harmony remote and it amazes me how she somehow seems to get the touch scren to places I've never seen before!

[markgoldie]

CBedo;

Nice post and I agree with everything you said in it and believe me, this is very rare for a post that long and for what I normally read on the site.

Personally, I am interested in automated handicapping for precisely the reasons you mentioned. I can't possibly do the standard handicapping and handle all the races I want to. I should say that I handle about $3m per year and am qualified for rebates on a "mini" whale basis. So my problem becomes one of needing lots of races or playing huge amounts into fewer races. I prefer the former because it gives me a "smoother" result without the crazy ups and downs.

When I use automated info coming from Bris, for example, I am constantly haunted by the desire to tweak their algorithm. And I firmly believe it can be tweaked with some consistent success. That's not to say that their Prime Power algorithm is bad because it's not. It's just that with everything that it must do, maybe they are not aggressively tweaking it themselves in all areas. This seems to be the case.

At any rate, one of the key areas in which their number seems to be tweakable is in the "returning from layoff" conundrum. I can do the tweaking to my satisfaction myself, but again, this puts me back into the "labor intensive" area which I am trying to avoid since I am already tweaking other things.

That being said, there are clearly many areas of possible tweaking that I ignore simply because I don't have the time or ability to tackle them. Watching replays and assessing trips, charting workout patterns, visually studying the horse, charting the odds' board, keying in on any daily speed bias of the track, and following jockey changes are just some of the things I don't do. Here, I have to hope that my lack of attention somehow "comes out in the wash" and either the law of averages saves me or the attendant greater odds for the stuff I missed somehow compensates me over the long haul. Would I like to be able to do all those things? Sure. But I can't and I realize that what I need to do today to win is far different from when I was young and sitting at the track immersed in the nuances of the single venue. Today that is impossible, not the least because of the unbettable short fields and garbage maiden races that many tracks are serving up so often. And also the pathetic handles at many tracks whereby a $1000 win move on a horse can crash the odds' board.

So, I ask questions to the computerized cappers to see where their tweaking is going. Maybe you'll be the guy who puts it all together.

Thanks.

Mark