This is hard machine learning problem. The task is to predict cryptographic SHA-256  hash output (10 first bits) using machine learning methods and without using detailed  knowledge of SHA-256 function. It is interesting to find out which methods can learn very complex non-linearities.

Input data is 10 dimensional vectors / strings of ASCII values (message) and output is 10 bits of y = SHA-256(message). Mean and standard deviation of SHA-256 bits are 0.50 meaning there are no easy statistics machine learning methods can use.

Have fun!