seed value

[ehsan-soe]

Hi,

I have trained a classifier and I am trying to do inference for a text file rather than a single input. I realized that I will get a different outputs if I reorder the examples in my file. and this is because of the seeds here. Does it really matter to generate all examples with seed 0?

[dathath]

You will get different samples if you change the seed, but that should be OK.

[ehsan-soe]

For some reasons that I cannot figure out yet, when I generate single samples vs when I generate using all inputs from a text file, the first gives better results. Do you have any guess?

[ehsan-soe]

@dathath I think I figure out the reason. In this regard, should the perturb and unperturbed generation happen together? The reason I am asking is when I comment out the function call for the unpert generation, the result of pert generation degrades.

[dathath]

Can you point out of the lines you are commenting out? As long as you're not disabling the KL/GM terms, it should be fine.