Skip to main content

Oops I didn't check stochasticity

Created on June 17|Last edited on June 21

Showing first 10 runs
05k10k15k20kStep1100
3 1.0000e-7
3 0.000001
3 0.00001
3 5.0000e-10
3 1.0000e-18
3 1.0000e-30
3 1.0000e-10
3 5.0000e-11
3 5.0000e-12
3 5.0000e-15
Run set
35

Ok, now let's see its performance on other datasets.

Run set
10

Let's see if I can't push those more complex datasets through. You really only see it retaining longer-term dependencies around the 0.7 loss mark, and the loss appears to rise after the 2.5k iteration mark, so I'll bet some more layers/smaller learning rates will break the barriers.
Hidden size at 512:

Run set
25

at 64:

Run set
5

at 32

Run set
0