Real world problem/Challenge for you who claim to be geniuses:

Problem:

>getting cucked by another bot farmer at clayrocks, he has two, one per each rock, they never change.

My solution so far:

Keep track of each attempt to mine, and each win for each rock.

compute try rate minus the win rate for both,
choose the best ratio for my next attempt.

How can I improve this? (I'm using java) and 10 + level deep nested finite state machines (rock selection is isolated to a single level of a machine), but I'm interested and motivated enough to implement some kind of predictive modelling/neural networks

*I'm somewhat limited with what I can do inside java and want to keep the cpu overhead down as much as possible, so no openCV.

c...could I gather a ton of data and then choose the rock based on a precomputed function (calculated by a neural network fed the win data from my farm)?