Quote:
Originally Posted by classhandicapper
Here's an interesting one.
Suppose I had a spreadsheet with 5 factors calculated inside my database.
1. Class Rating
2. Speed Rating
3. Consistency rating
4. Running Style (0-5 with 0 being a pure speed and 5 being a deep closer)
5. Projected Pace Pressure this Race (0-30 rating with 0 being no speed and 30 being a race loaded with speed)
How would you approach testing how to weight each of those factors so the spreadsheet would understand that as the Projected Pace Pressure changes, the optimal Running Style and it's importance would also change?
Maybe that needs to be done separately?
|
A couple of different ways come to mind. For example with the first three factors of class, speed and consistency you could take the rank approach similar to the first example spreadsheet with a 1-4 rank and 'other' for the rest. Then for the running style and projected pace piece of this I'd probably start by trying to handle it using a pair-up. I'd probably start by looking into 'binning' the projected pace total of 0-30 ... throwing out some aircode here IF(AND(RS =0,PP>=0,PP<=5)),then point to one of the cells solver will change...
IF(AND(RS =0,PP>=6,PP<=11)) then point to the next cell down from that one etc...
so that will be five? bins for each running style 6x5 = 30, it's a lot of combinations but anyway that's one way to do something with it. Maybe with testing you'll find that only the extremes really matter, i.e. maybe larger bins and / or fewer running styles actually matter etc. In theory they all matter I know but you're projecting pace and we all know that's bigtime hit and miss, so what might actually make money on a sample could be the very opposite of what people expect.
Another approach could be for example to find the point value of a pairing. So we have our 30 pairings and we have our speed or our class, how are those scaled? Let's say 'speed' for you is a BSF, we take the BSF and we use solver to adjust that by the solver values of the pairings, that way we optimize on winners based upon BSF but the BSF is being adjusted in another column by the paired value and that adjusted BSF first ranked is what we bet.
So that's at least two ways for you to look at it... and maybe neither are good ways but hopefully I've provided some food for thought.