Quote: Originally posted by vick on April 16, 2004
i used neuarl netwoks years back on the lottos = not much sucess, byt mabye today's faster computers can work things differently with it?
After reading a great book on pattern matching NNs and months of ehxaustive web research and toying with excel and stand-alone demos I have come to the conclusion that a NN as designeed today cannot help with the lottery... it is an attempt to model the human brain, and we already have better abstract pattern recognition than the fastest cray supercomputer...
I did take away 3 things from the research that could possibly be incorporated into a predictive engine...
1. Fuzzy Logic... Implemented as a simple nested IF...THEN statement. THe classical IF...Then or If...Then...Else logic switch traditionally has 2 outcomes
>> If SomeCondition = True Then SomeAction, Else If SomeCondition = False Then SomeOther Action
Example, If X < 10 then Y=1, Else Y=0 Has only 2 outcomes
Fuzzy Logic can be implemented as a nested IF statement...
>> If SomeCondition = WithinSomeRange Then Some Action, Else If SomeCondition = Within the next SegmentOfTheRange Then SomeOtherAction, Else If SomeCondition = WithinNRange Then NthAction.
Example, If X<10 then Y=1, Else If X <8 then Y=0.8, Else If X <6 then Y=0.6, Else If X <4 Then Y=0.4, Else If X<2 then Y=0.2, Else Y=0 now Y has a range of 6 outcomes based on X.
2. Weights... having a set of statistics for the next draw and weighting each based on past performance. Let's say you have 3 stats, Stat B perfomed better than Stat A, which did better than Stat C... but you want to incorporate each, giving weight to the better performers...
Result = INT(AVERAGE(((B*3)+(A*2)+C)/6)) <<< as done in Excel
3. Long Term Memory (Full Draw History) and Short Term Memory (Last N Draws)... combining these with weights should marry long term performance with short term trends... but at what value of N? Guru uses 3, I have seen better results with 5...