Colin F,
Dang, I had a reply all worked up and somehow it never posted...
Anyway. What I had said, was that a test was run last year during the month of March. Several users posted predictions from the software, along with Random numbers generated by Random.ORG. I myself only participated in the trial for the first week.
The results brought out two distinct issues.
- Random numbers out performed the user designed stacks in that test example.
- That the user designed stacks was only a small set of possible prediction schemes possible from WINHunter.
There is discussions of a full blown prediction test of ALL lottery software to date going on in the RGL newsgroup. This prediction test would entail a detailed explanation of how to repeat the prediction tests with said software. These prediction tests would be run against several lottery histories, and would be performed against past historical data.
The prediction sets would be of raw numbers only, No wheels. This would prove how well the program performs at choosing the base set of numbers, since a wheel is only as good as the numbers fed into it. As far as the results of using Random numbers and WINhunter performance, the tests speak for themselves. But again, in regards to WINHunter, the test only showed how well THOSE stacks performed.
My goals in regard to Performance, is to find a stack that can consistenyl predict small sets of numbers, and still "Trap" the winning combinations. Using a stack I built over a year ago, you can see the past performance of this stacck against the entire Florida lottery history (https://www.lotterypost.com/thread/81341). This stack may not "Consistently" (as in on a weekly basis) predict sufficient numbers to wheel, let alone predict enough matches. The goal in this optimization, was to capture the lowest number of balls drawn that generated a matching Jackpot Win. This tack produced the August 28th, 2002 prediction. Mind you, this prediction was a month old at the time (it was September when this stack was created). But look at the draw #15 in the rundown list. That is the December 20th, 2003 prediction. Those are the numbers WINHunter would have predicted for the draw that day. Is this useful or beneficial? You tell me....
You can't think of WINHunter as a "system", but you must think of it as a System Designer. The WINHunter project is in no way complete, in fact, it has only just begun. There are tons of analysis programs out there, and tons of wheeling programs. WINHunter brings a totally new set of features to the table. It shows concepts not ever used before in Lottery software. Think of WINHunter as a proving ground.
Let me tell you about an example function that I was allowed to convert into a WINHunter function. An RGL regular member described a function, which used a fixed set of values to use different groups of past draw history to score numbers based on which group they appeared in. His method was origionally performed by hand. It took maybe 2 days to complete the conversion of his function into a WINHunter processor. This Processor enabled the user to test the function in any way s/he saw fit, against whatever amount of past history s/he wanted. By incorporating this function as a processor, WINHunter was able to read the configuration options and thus perform Optimizations with this Processor. Optimizer results could be viewed and stack designs saved for further optimizations.
One of the interesting things to note about this processor, is that WINHunter would find a slightly better performance gain by adjusting some values of the Processor from the inventor's defaults, depending upon the history being tested with.