Skip to main content

torn apart

a look at what the rnn is doing step-by-step
Created on June 13|Last edited on June 17

Showing first 10 runs
050100150200250Step11e+20
Run set
802



Run set
5703



Run set
9



Run set
90



Run set
10

Did I get the deets right? It really seems like it may be the candecay value of 0.99 that is the secret sauce, at least for getting results that pass the basic test of a repeated sequence of length 2. Let's see if it's that value that really fixes things.

Run set
68

is it the number of layers??

Run set
25

Of course, my initial test had a curve upwards near the end. I think a different learning rate helped with that.

Run set
53