JayR

If we want an approximately similar distribution to the old code and if my binomial distribution calculations are correct for a value like S=10 the code would look something like...

``` On the MNIST dataset using the integer version with r = random() % (s + 4);... TimeBefore TimeAfter TestBefore TestAfter 378.1 164.6 93.9 94.0 232.4 126.6 95.1 95.1 203.3...

Edited: The original optimized version seemed to converge slower toward epoch 6 for MNIST. I switched to s=11 and then used this code and it matches or exceeds for the...