The number of draws of game history to use is determined by regression testing. These type of spreadsheets take along time to type in. 1 sheet tells you how many draws of history to use to find out how many games back the most frequent number or numbers came out. For example I start with 7 games of history, then test one game back for up to a year of game history. This tells me how many times the most frequent number or numbers hit the next game using 7 games of draw history. Then I do the same thing for 8,9,10 etc.,etc. until the sheet has hit it's max columns (256 for 32 bit). I then look for the number of game history that hit the most. in the fewest number of games. In my own tests I find this changes over time as the game history changes. in NC Pick 3 right now 42 draws of history hit the most at about 34 percent of the time. That does not sound all that great at 34 percent but it's a more solid degree of certainty to work with than just guessing / hit or miss. I do this also for sums and am working on doing the same thing for other settings in nnlp. However I am finding that using nnlp may not necessarily produce better results than simpler methods. I also look at regular sums ... I create a sheet that has a list of all the sums and their associated box number sub lists. separating mixed and double numbers Showing me the sums that hit most frequently first. Then listing the box numbers for each sum sorted by their most frequent deltas, which correlates with how many times a box number came out in the last year.
