I know that with regression it is possible to generate a mathematical function based on experimental data. I was wondering, maybe there is a way to generate something more complex. I have a single player logical game which is hard to solve:
There are 37 glass globes in holes:
|o|o|o| |o|o|o|o|o| |o|o|o|o|o|o|o| |o|o|o|o|o|o|o| |o|o|o|o|o|o|o| |o|o|o|o|o| |o|o|o|
You have to remove one globe at the beginning (I started with the middle one):
|o|o|o| |o|o|o|o|o| |o|o|o|o|o|o|o| |o|o|o| |a|b|o| |o|o|o|o|o|c|o| |o|o|o|o|d| |o|o|o|
After that you can step over a single globe in horizonatal or vertical direction (I stepped here with "b" over "a", so I removed "a" from the playground):
|o|o|o| |o|o|o|o|o| |o|o|o|o|o|o|o| |o|o|o|b| | |o| |o|o|o|o|o|c|o| |o|o|o|o|d| |o|o|o|
Now I stepped again (with "d" over "c"):
|o|o|o| |o|o|o|o|o| |o|o|o|o|o|o|o| |o|o|o|b| |d|o| |o|o|o|o|o| |o| |o|o|o|o| | |o|o|o|
The aim of the game that only a single globe should remain. Now that is extremely hard to achieve, I have less than 1% success rate and I don't have the winning strategy.
Is it possible to generate an algorithm, which can solve the problem by many different scenarios (so which has a strategy)?
I was thinking on genetic algorithms or neural networks, so something which can learn the basics from me and after that develop itself by gaming more, but I have no experience with these methods, so I don't know where to start, and which one would work. I am aware that it is possible to solve the problem manually, but from that I would not learn anything about artificial learning...