Paceline Selection Algorithms

[Jeff P]

I've been following the back and forth over Paceline Selection between Light, Doug, HeadHawg, and Handi in the Handifast thread... Some good points raised on both sides...

Pacelines... To pick or not to pick?

On the one hand you can create algorithms to evaluate, score, and weight every running line... and come up with pretty good compound numbers that test out nicely roi-wise.

On the other hand there is something to be said for allowing the expert handicapper to pick his (or her) own lines.

In my own R&D I've always leaned more towards creating compound numbers... as I've never been able to (meaning me personally) pick lines in a way that outperforms (using flat win bet roi in large data samples as the benchmark for comparing two different approaches) what I've been able to achieve using a compound number approach. I've written a few paceline selection algorithms myself... but let's save that topic for later... I'm hoping to get some input first.

There are a lot of intelligent posters on this board. I'm curious about your thoughts...

Pacelines... To Pick... or Not To Pick?



-jp

.

[Red Knave]

Jeff
This seems to me to be one of several areas of handicapping where art is still more useful than science. And, same as you, I know I'm no artist.
For those who can visualize an entire race, picking a representative paceline seems obvious. For me, not so much. Dick Schmidt, when he edited the Sartin Follow Up, said he always used the last line except in the case of a nuclear attack (I'm paraphrasing, but words to that effect). Jim Bradshaw just intuited the correct line because he knew what was going to happen during the running of the race.
In my handicapping, lazy and tentative as it is, I prefer NOT to pick pacelines and use software that gives me a composite view of each horse that I can use to compare them one to one as well as one to the entire field. I feel a lot more comfortable with that than if I had just one or two hopefully representative lines.

[Light]

Jeff

This is a vast subject,but let me tell you why their paceline method is flawed using the 3 basic scenarios often encountered.

1) Many horses drop in class. Used to run for graded stakes and are now in claiming ranks. Does it make sense to you to use a race a few races back where the horse ran a 90 vs stake competition and now cant reach an 80 with claimers? Not to me. The owner and trainer have already said to the readers of the pp's,"my horse cannot do that 90 anymore and he cant even do an 80 vs claimers so I'm dropping him some more". Yet people who average x number of pacelines are ignoring this really obvious form factor and using the horse's 90 when he used to have 4 legs and now uses a walker.

2) Nobody has ever come up with a turf to dirt equivalent chart. Everyone knows its apples and oranges. But those who average x number of pacelines back ignore the fallacy of using incongrous surface change figs.

3) Bris distance figs are not the same for a sprinter and a router. A 100 in a sprint and a 100 in a route are not equal. Which is better depends on todays distance. You really need to make an asterisk on distance switchers and apply a totally seperate criteria to see if a sprinter can stretch out or a router can sprint.

[Warren Henry]

Originally Posted by Light

Jeff

This is a vast subject,but let me tell you why their paceline method is flawed using the 3 basic scenarios often encountered.

1) Many horses drop in class. Used to run for graded stakes and are now in claiming ranks. Does it make sense to you to use a race a few races back where the horse ran a 90 vs stake competition and now cant reach an 80 with claimers? Not to me. The owner and trainer have already said to the readers of the pp's,"my horse cannot do that 90 anymore and he cant even do an 80 vs claimers so I'm dropping him some more". Yet people who average x number of pacelines are ignoring this really obvious form factor and using the horse's 90 when he used to have 4 legs and now uses a walker.

2) Nobody has ever come up with a turf to dirt equivalent chart. Everyone knows its apples and oranges. But those who average x number of pacelines back ignore the fallacy of using incongrous surface change figs.

3) Bris distance figs are not the same for a sprinter and a router. A 100 in a sprint and a 100 in a route are not equal. Which is better depends on todays distance. You really need to make an asterisk on distance switchers and apply a totally seperate criteria to see if a sprinter can stretch out or a router can sprint.

So, you have identified three areas of concern regarding paceline selection, but I don't see any suggestion for overcoming the problem. Enlighten us.

[sjk]

I have had good luck with compound numbers (although I have no actual number because it is all internal to the program processing). Makes good sense to me to use all of the relevant information you can.

[raybo]

AllData PPs will have an auto-paceline select method, with 14 selection criteria defaults that the user has options for each one. It also has these 14 criteria grouped into 6 categories which the user ranks (1 to 6) in order of importance. The program will look for perfect matches first then proceed to drop the least important category and re-search for matches, then the 2nd least important, etc., etc..

The user also has the option of overriding any or all of the auto-selected pacelines by checking a cell next to his preferred paceline.

The final selections will then appear in a summary with only the selected paceline for each horse

[Tom]

I use a couple of the auto-modes in HTR, but I always look at them and change what I don't agree with.

[BillW]

Originally Posted by Jeff P

On the one hand you can create algorithms to evaluate, score, and weight every running line... and come up with pretty good compound numbers that test out nicely roi-wise.

.

I generate/evaluate both .

[douglasw32]

I think if you are going to use pacelines, become an artist.
It is far to subjective for a computer.

[Light]

Red Knave and douglasw32 both refer to selecting your own paceline as an art. I would say that's true in only a small number of cases. In the vast majority of cases its a no brainer which paceline to pick.

[raybo]

Originally Posted by douglasw32

I think if you are going to use pacelines, become an artist.
It is far to subjective for a computer.

I agree that the auto-selection process should be verified by the user and overridden when it is obviously in error. If one uses a thorough approach, in programming the auto-selection criteria and decision matrix, most of the time it will yield the same paceline the user would have chosen manually. Auto-select saves handicapping time and helps prevent user errors by ensuring the fundamental selection criteria and method remains consistent.

[raybo]

Originally Posted by Light

Red Knave and douglasw32 both refer to selecting your own paceline as an art. I would say that's true in only a small number of cases. In the vast majority of cases its a no brainer which paceline to pick.

The only time it's a "no-brainer" is when there is only one paceline, or none. However, after that, the brain better be working overtime.

[douglasw32]

Originally Posted by raybo

I agree that the auto-selection process should be verified by the user and overridden when it is obviously in error. If one uses a thorough approach, in programming the auto-selection criteria and decision matrix, most of the time it will yield the same paceline the user would have chosen manually. Auto-select saves handicapping time and helps prevent user errors by ensuring the fundamental selection criteria and method remains consistent.

Yes that is a very good idea, make sure with the computer the human does not mess it up but then as the human know when to tell the computer to take a hike.

[headhawg]

Originally Posted by raybo

The only time it's a "no-brainer" is when there is only one paceline, or none. However, after that, the brain better be working overtime.

That's the same way that I think about it. You could use the "good enough" method which basically what Mitchell said about using the last paceline 90% of the time. It's kind of like using TPR or Bris figs. They just need to be good enough because we're not trying to land a shuttle on mars. But the implication with that approach is that there is something else in one's handicapping that supplements the paceline. That is, the paceline is not the be all, end all part of handicapping, but merely one step in a longer -- and hopefully more revealing -- process.

On the other hand, if one feels that picking pacelines is the way, the truth, and the light, then you better be damn good at it. And I am not convinced that there is an auto-selection method out there that is decidedly better than a compound method that Jeff P and sjk may use. Obviously without seeing the algorithms I am in no position to evaluate them. They could be complex, but that doesn't necessarily make them useful over and above using the last line which would be simpler. Or it may not even be better than the averaging method that Handifast uses.

And as I have mentioned in the other thread I don't have anything against paceline selection either manually or automagically (my own software does it). I just don't think that a program has to do it that way, just like I don't think that a program that doesn't include trainer stats as a significant selection factor is fatally flawed. It's utter nonsense to think that there is only one way to design a handicapping program.

[LottaKash]

For me, I will almost always use the most recent paceline, provided that there is a genuine excuse for a loss or a poor performance off of that line, than I will go back as far as 4-back if the first 3-are excusable...after that, a crystal ball may be needed...(I don't use software, tho)

Can a software program know when a running-line may be excused ?

best.