GA-Fitness Scoring for Betting Strategies

[podonne]

I'm trying to develop a relaible fitness score for comparing different betting strategies.

So I've been playing around with genetic algorithms to develop betting systems for a while now, but the same problem keeps popping up, so I thought I could put this discussion here.

In order to use a genertic algorithm, you need a high quality method of developing a "fitness score" to evaluate generation members against each other. There has to be one score, a single number, and a single way of evaluating it against other scores (greater than, less than, closest to 1.0, etc...). I wonder what type of score would work for betting strategies? Surely not just ROI or Hit%, because of the outlier longshots, but maybe a more systematic and comphrensive method?

Any thoughts?

[betovernetcapper]

Impact Value?

http://www.netcapper.com/TrackTractsArchive/TT010223.htm

[Robert Fischer]

I'm trying to develop a relaible fitness score for comparing different betting strategies.

So I've been playing around with genetic algorithms to develop betting systems for a while now, but the same problem keeps popping up, so I thought I could put this discussion here.

In order to use a genertic algorithm, you need a high quality method of developing a "fitness score" to evaluate generation members against each other. There has to be one score, a single number, and a single way of evaluating it against other scores (greater than, less than, closest to 1.0, etc...). I wonder what type of score would work for betting strategies? Surely not just ROI or Hit%, because of the outlier longshots, but maybe a more systematic and comphrensive method?

Any thoughts?


Sounds very interesting.

I really would have to get comfortable with genetic algorithms to give the best input.

Hit % is obviously not the choice. Why not ROI? Are you saying that if someone hit a "lucky" longshot, it could skew the results?

[Overlay]

For a single statistic, what about $NET (average return for every $2.00 wagered)? Wouldn't that take both winning and losing bets into account, be sensitive to changes in hit rate and mutuel payoffs, and also smooth out the effects of outliers?

[Dave Schwartz]

I create several variable fields, each to address my priorities.


Example:

Hit Rate (positive)
Hit Rate (negative)
$Net (positive)
$Net (negative)
Optimum Bet (advantage / odds)
Avg Wager Size(larger)
Avg Wager Size(smaller)
Avg Payoff Size (smaller)
Avg Payoff Size (larger)
Pool Impact Value (larger)
Pool Impact Value (smallerer)

Each of these fields contains a second field which contains the "base." Thus, I can (for example) penalize any $net that is under (say) $1.85 using one weight and credit any $net over (say) $2.00 using another.

Hope this helps.


Regards,
Dave Schwartz

[Jake]

I create several variable fields, each to address my priorities.


Example:

Hit Rate (positive)
Hit Rate (negative)
$Net (positive)
$Net (negative)
Optimum Bet (advantage / odds)
Avg Wager Size(larger)
Avg Wager Size(smaller)
Avg Payoff Size (smaller)
Avg Payoff Size (larger)
Pool Impact Value (larger)
Pool Impact Value (smallerer)

Each of these fields contains a second field which contains the "base." Thus, I can (for example) penalize any $net that is under (say) $1.85 using one weight and credit any $net over (say) $2.00 using another.

Hope this helps.


Regards,
Dave Schwartz

Dave has worked with GA's for quite a long time. He was be the person to talk to here.

Jake

[osophy_junkie]

An ideal fitness function will take the ROI of a wagering method and number of races played into account. It will also be lenient towards negative ROI's during the first few generations as to build models that are valid for a greater number of races.