Comment by zaptheimpaler

How is it possible to make a general coefficient of correlation that works for any non-linear relationship? Say if y=sha256(x), doesn't that mean y is a predictable function of x, but its statistically impossible to tell from looking at inputs/outputs alone?