search

aigamedev / scikit-neuralnetwork

Deep neural networks without the learning cliff! Classifiers and regressors compatible with scikit-learn.
BSD 3-Clause "New" or "Revised" License
1.2k stars 221 forks source link

Seeding Random dropout #203

Open jimmyroyer opened 8 years ago

jimmyroyer commented 8 years ago

Hello, is it possible to seed (initialize) the random dropout regularization in the MLP classifier? Thanks

alexjc commented 8 years ago

I presume you tried seed as constructor argument and it's not deterministic?

jimmyroyer commented 8 years ago

Seed works but it seeds the whole function. For instance if I use a randomized search for the meta parameters it will seed the searched parameters as well as the dropouts and the weights. I was wondering if it is possible to only seed the random dropout part of the classifier.

alexjc commented 8 years ago

No, it's not currently possible. Could you share why you need dropout-only seed?

jimmyroyer commented 8 years ago

I want to use the seed of the weights and the randomized search as another "hyper" parameter.

alexjc commented 8 years ago

I'm not sure I would support dropout as separate seed, it shouldn't make any difference as the iterations increase. More important is to expose the weight initialization strategy!