DBM - Hinton & Salakhutdinov paper #1239
Conversation
…DME update, some files moved in other directories
augmented datasets. corrections on fast test script
binarized test set in dbm softmax added
|
@carloderamo Can you rebase and squash this into a single commit? As this is all concerning DBM stuff that nobody knows anything about other than @goodfeli I'm going to punt the review to him. |
| from pylearn2.utils import serial | ||
|
|
||
|
|
||
| class MNIST_AUGMENTED(DenseDesignMatrix): |
There was a problem hiding this comment.
You should not need a new Dataset object to do this.
|
Since I've requested relatively major changes based on the small parts I have reviewed, I'm going to quit reviewing early and will review the rest if / when the first changes are made. |
|
Currently I have a lot of work to be done for completing my MSc thesis on Carlo D'Eramo Since I've requested relatively major changes based on the small parts I — |
Pylearn2 implementation of the whole procedure to train a DBM described by Hinton & Salakhutdinov in "An Efficient Learning Procedure For Deep Boltzmann Machines".
This pr adds a fast version for the test script to be run on travis together with some changes to the format and the readme file. I also moved augment_input from datasets to /scripts/dbm.