I have been working on programs to play lotteries since last July, when I started prototyping. In the last day, I have been testing performance with a range of numbers for playing Cash Five (Texas). I had always assumed that generating combinations from a greater number of numbers was superior to a smaller number. I proved to myself today that is incorrect. Playing as few as 11 hot numbers was better than playing 33 hot numbers. This is using a "full wheel" with random selection of combinations to play. One of my tweaks was to randomly select small blocks of combinations, and that often is better than a file of single random selections.
Besides using "hot numbers", I now use a "full wheel", and gather statistics about what lines from the wheel to play. One thing that I discovered over the last nine months is that some combinations from wheels will win more often than others. So, the natural thing to do is to play those lines and ignore the ones that never win or that don't win very often.
I have seen how well my "Pick 3" program has done over the last few weeks. I use a combination of "hot numbers" and "positional analysis" to generate picks for the day and night games. I have been unhappy with my existing approach for playing games, particularly "Cash 5". I redesigned the "Cash 5" program to use "hot numbers" and got immediate results. I have spent the last four days refining that program. Yesterday, I adapted that program to play "Lotto Texas". "Lotto Texas" is a much different game, because of the erratic winning numbers, the six numbers per combination, and the 54 numbers played. One result of that is that on many days, my program will win nothing (I run against the history for testing purposes). However, my program is good enough to do well often enough to be profitable (it looks like). I am now playing "Pick 3" day and night, "Cash Five", and "Lotto Texas" (twice a week).