Light (Single Paceline Selection)

[douglasw32]

Light-

I modified a program based on the hadifast code (the lil Brother)
to work selecting a single paceline from the last 3 races, based on distance.

I have to admit, you are correct it improved the selection ranking.

I did find that when I refine it more however, go to only 1/2f, include only fqst dirt, etc.

It starts to do as Jeff P suggests.

It did however Improve the program as simplistic as the selection method is.

And although the hadifast camp see's no reason to include it in our tool, using handifast in conjunction with a program that does do it, really , really improves both of them.

I will post the improved one to the hosting server soon for anyones review/comments.

[lsosa54]

Doug: My opinion is that Handifast's value is that it focuses beyond the last 3 races, which the public basically ignores, as well as, some of the unique ratings. I think focusing on a single paceline within the last 3 might bring you a more orderly ranking in those races which may be less chaotic but it would reduce the mutuel value of some of the winners Handifast points to.

I prefer to use Handifast as is with some logic as to when the 6 race averaging may give faulty results and then use those contenders in a paceline oriented program which may be better suited for narrowing down the contention.

However, given some oi the results I've seen recently, simply playing the "value" in the Handifast fair odds line may be the way to go, although I'm still trying to determine whether a 5th or 6th fair odds choice say at 10 or 11-1 going off at 20-1 is "value" or just an inaccurate line, as the horse has little chance of winning. Just check out the GP results from yesterday.

[Handiman]

Doug,

Can you be more specific in how it improved the program and what the parameters were that were involved in the pace selection.

I don't think the guys when sitting around the ole campfire mind an open discussion about possible improvement albeit Pace Line selection and inclusion, it's just we would want some civility from the main motivating speaker.

We all have a great deal of respect for you and your intentions. And on top of that this program is not close to it's half life, let alone done. So please kick down more details if you would.

Handi

[douglasw32]

Originally Posted by lsosa54

Doug: My opinion is that Handifast's value is that it focuses beyond the last 3 races, which the public basically ignores, as well as, some of the unique ratings. I think focusing on a single paceline within the last 3 might bring you a more orderly ranking in those races which may be less chaotic but it would reduce the mutuel value of some of the winners Handifast points to.

I totally agree, and that is why I started coding a second program (and I was bored) to do so, not to muck up what works in handifast but to create something new from the open source code and prove that there are more than one way to skin a cat so to say.

[Viruss]

I have been looking though some old files using Handifast.

MNR only

The winner is one of the top 3 ranked horses out of the S1-2-3 about 80% of the time. So I for one can't wait to check it out.


Earl J

P.S. No i never checked the distance.

[Light]

Doug

I want to make it clear that my pl selection criteria is based on class and surface as well as distance. I realize you may be just using distance as a convenient method to code for selecting a single pl, but I dont want any misunderstandings. I appreciate your interest and openness.

[raybo]

For once I agree with Light, I think?

If you are going to select only one paceline, you can't use just distance and you can't restrict to last 3, or last 4.

[douglasw32]

Originally Posted by Handiman

Doug,
So please kick down more details if you would.

Handi

You would have to look at the code, I took a totally different (well sort of) approach.

It cycles through races, first it creates an adjusted speed figure looking only at the last race, then at the last 2, then the last 3.

It looks at pace from the last 6, finds the best one.

Averages it into the 3 totals produced.

Adds in the Form Points, and the distance points.

averages all of them into a power rating, and then divides it by ten
bringing it into a single number .
11,10,9,8 etc.

Then from the last 3 races it looks for a match 1f plus or minus in distance and takes the best bris speed figure.

if none it uses the last race unless one back is higher.

adds the power number into it, then alla Randy Giles www.paceappraiser.com/
ideas it moves up any single running style, lone E only one S etc.

The modified speed figure per race is something I have played with a long time, it takes the best pace e1/e2/lp how many lengths the horse closed, how much Early Speed he had, the number of horses ran against, the class of the race based on the top3 finishers, weight carried and the number of horses in the race and the horses speed figure it is then normalized into s new modified speed figure.

Sort of the FAST idea but for each race not the average of races.

HArd to explain without seeing the code I guess but that's an idea.

I will upload the newest version of it that outperforms the one I have posted, when I upload the new handifast, so both of them are at there newest version and can be used to compliment one another.

[douglasw32]

Originally Posted by Light

Doug

I want to make it clear that my pl selection criteria is based on class and surface as well as distance. I realize you may be just using distance as a convenient method to code for selecting a single pl, but I dont want any misunderstandings. I appreciate your interest and openness.

Okay thanks, you are correct the distance was just the first piece I threw in because as you pointed out it is totally ignored in the averages in the other program and it was convenient to do so.

I was going to go deeper and add in the surface and class but started getting really solid results so I am just holding off for another day.

I guess 1/3 of the paceline criteria is covered instead of none.

But I think I am seeing the Jeff P, Scenario mentioned in the other thread.

That for lack of better terms, it does not make much of a difference what PL selection method you use as long as you are using one ???

[RichieP]

Originally Posted by lsosa54

However, given some oi the results I've seen recently, simply playing the "value" in the Handifast fair odds line may be the way to go, although I'm still trying to determine whether a 5th or 6th fair odds choice say at 10 or 11-1 going off at 20-1 is "value" or just an inaccurate line, as the horse has little chance of winning.

Hi Lou
Try tinkering with field size and random odds for that particular races # of runners as a way to determine how deep to go on your odds line.

Think it goes like this:
12 horses = random odds =11/1 (consider all runners at those odds or less on YOUR line)

10 horses = random odds = 9/1 (consider all runners at those odds or less on YOUR line)

8 horses = random odds = 7/1 (same as above etc)

This might help clear some fog and also treat each race as the unique entity it is.

Knock em dead and best of luck/skill

[lsosa54]

Thanks for the idea, Rich. Just got home from work so I'm still in a fog!

You may have something there with field size. In keeping records for the BRIS Prime Power stats since the mid 90's, field size was the biggest factor in their effectiveness for contender selection.

For example, in 6 horse fields, the winner would come out of the top 4 ranked horses close to 90% of the time on the dirt and the top 2 arund 50%. Once you got the field size up to 9+, you'd be looking at 72% for routes and about 74% for sprints in the top 4. The turf field size differentials were even bigger, esp. for routes. Those extra 3+ horses make things a lot tougher.

I will definitely work your field size idea in.

[douglasw32]

Agreed. It fails most of the time with maidens , 2yo and most turf races...

[douglasw32]

Oh CRAP !!! this whole being a moderator thing gives new buttons to the poster... LIGHT ! I just deleted the post I was responding to, if you could add it back into the thread...

You were pointing out how it would depend on the type of race and I responded to you.

[PaceAdvantage]

Originally Posted by douglasw32

Oh CRAP !!! this whole being a moderator thing gives new buttons to the poster... LIGHT ! I just deleted the post I was responding to, if you could add it back into the thread...

Yeah...you have to be a tad more careful now...

[Light]

Originally Posted by douglasw32

Oh CRAP !!! LIGHT ! I just deleted the post

I'm sure others are not that upset about it.

What I said was that I agreed with you about the negligible difference between PL selection methods in certain races,mainly older,more experienced types that usually run the same class,distance and surface.

I think there are several instances where the single Pl method beats the multi pl method.

1) Horses improving/declining in form. For example a trainer change, medication change (published or behind the scenes) or equipment change may result in a 180 degrees form reversal of a horse's form and using the pre change races will give an unrealistic representation of the horse's ability today.

2) Lightly races horses who have only a few lifetime races. For example a horse with only 2 or 3 PP's in its form. First race was a 65 SR, second was a 90,third was a 90. Single PL would give this horse a 90. Averaging would be sticking to a rule for the rule's sake rather than seeing the horse's obvious ability.

3) Surface changes including on/off tracks. One of the most interesting I had recently was a horse who ran on a sloppy track four races back that had a 10 point higher SR in that race than his last three races on a fast track,all at the same distance,track and class. Today's race was a sloppy track. He went off at 29-1 and came flying to just miss beating the favorite. Pinpointing this surface change with a single pl selection was the only way to identify this bomb via pl selection. Averaging would have dismissed him.

And I also had mentioned that you can create a model of winners more accurately with a single pl format rather than a multi. For example, I want to know what type of single PL horses come out of that win a certain claiming sprint. Its easy to see when databasing. I'm only dealing with 1 race the horse came out of. But if I want to do the same with the multi,I have to find a horse that has all the pl races that were chosen to form the average rating this past winner earned. An impossible task.