Program for number prediction - using neural network for lottery

I would order a program that would use a deep neural network to predict a numbers from a lottery. 15 numbers out of 25 (1 to 25). Based on history results stored (see attached file). I need the program to be easy to use, and to allow me to add new historical results and alter the results based on new data I add to it. I need the program to run under windows and use GPU (nvidia CUDA) for faster computing. There are 3 draws in a week, always 15 numbers (mondays, wednesday and friday). As we choose at least 15 numbers (in 25 in total), we can "predict" at least 5 numbers always (with any combination). The same combination, as I think, never be drawed 2 times in history. Is important that I could choose 15, 16, 17 or even 18 numbers. If possible, I want that system training find patterns to predict testing with previous contest. Like that: use 100 first combinations and try hard to get 101 (testing with really 101 result), then do the same to find 102 and go on and on until the last result (1785). Doing that with saving the way to get results and learn the "how to" solve this aleatory problem.