search

SMART-Lab / smartlearner

SMART-Learner is a machine learning library built with researchers in mind.
BSD 3-Clause "New" or "Revised" License
10 stars 4 forks source link

Added a GradientNoise direction modifier #65

Closed ppoulin91 closed 8 years ago

MarcCote commented 8 years ago

Seems nice, can't wait to try it :). It misses the save/load functions so that training can be resumed. Edit: Check PR #62 for some examples. Let me know if you need help.

ppoulin91 commented 8 years ago

I'll have a look at that

MarcCote commented 8 years ago

So, from what I see, you'll need to save: t, std and _srng. The first two should be pretty straight forward. The randomstream, that's something else. I don't know if there any shared state you could saved. @mgermain I know you've played with that a bit, do you have any idea?

mgermain commented 8 years ago

I did tha in https://github.com/mgermain/MADE/blob/master/MADE/mask_generator.py and if I remember correctly you have to save the seed and the rstate.

On Fri, Jan 29, 2016 at 2:28 PM, Marc-Alexandre Côté < notifications@github.com> wrote:

So, from what I see, you'll need to save: t, std and _srng. The first two should be pretty straight forward. The randomstream, that's something else. I don't know if there any shared state you could saved. @mgermain https://github.com/mgermain I know you played with that a bit, do you have an idea?

— Reply to this email directly or view it on GitHub https://github.com/SMART-Lab/smartlearner/pull/65#issuecomment-176925635 .

ASalvail commented 8 years ago

I'll short circuit all this discussion by saying the save/load features are not yet ready to be merged and I need this before it'll be.

Once it's more stable in #62, feel free to make a PR against Marc's branch to include the save/load facility to this class.