Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Feb 1, 2010

• typo and bigger batches

  jaberg

  committed
  29f63bf

  Copy full SHA for 29f63bf

Commits on Jan 29, 2010

• partial rewrite of logistic_sgd to use a shared dataset. This makes the GPU a lot faster

  jaberg

  committed
  e2bfb2c

  Copy full SHA for e2bfb2c

• Account for new location of downsample outside of sandbox.

  

  Pascal Lamblin

  committed
  a3d5cb7

  Copy full SHA for a3d5cb7

• Merge branch 'master' of [email protected]:lisa-lab/DeepLearningTutorials

  Show description for 02c6539
  gdesjardins

  committed
  02c6539

  Copy full SHA for 02c6539

• changed.from theano.floatX.xmatrix to T.matrix

  gdesjardins

  committed
  3c328d1

  Copy full SHA for 3c328d1

Commits on Jan 28, 2010

• print only each 100 iteration.

  nouiz

  committed
  5fa898e

  Copy full SHA for 5fa898e

• fix following change to floatX

  nouiz

  committed
  21065c8

  Copy full SHA for 21065c8

• Fixed signed bug in weight initialization

  pascanur

  committed
  4256877

  Copy full SHA for 4256877

Commits on Jan 27, 2010

• A small bug in both version of stacked denoising autoencoders fixed

  pascanur

  committed
  d068e5d

  Copy full SHA for d068e5d

Commits on Jan 25, 2010

• updates hyperparameters of lenet network

  gdesjardins

  committed
  9525863

  Copy full SHA for 9525863

• made changes suggested by Yoshua

  Show description for 865ed2c
  gdesjardins

  committed
  865ed2c

  Copy full SHA for 865ed2c

Commits on Jan 22, 2010

• almost done convolutional net tutorial

  gdesjardins

  committed
  3ee9eaf

  Copy full SHA for 3ee9eaf

• Merge branch 'master' of [email protected]:lisa-lab/DeepLearningTutorials

  pascanur

  committed
  10a2de6

  Copy full SHA for 10a2de6

• Fixed a few bugs and defined a SdA with variable size

  pascanur

  committed
  c9e99a6

  Copy full SHA for c9e99a6

• latest updates to convolutional code

  Show description for e0e5e5d
  gdesjardins

  committed
  e0e5e5d

  Copy full SHA for e0e5e5d

Commits on Jan 21, 2010

• dae.py was replaced by SdA.py

  pascanur

  committed
  b3f50b7

  Copy full SHA for b3f50b7

• SdA.py implements a stacked de-noising autoencoder in the tutorial style. I think DBN/SdA are a bit different from the rest, so if anybody has the time to read through it and give some feedback I w…

  Show description for 76f20c3
  pascanur

  committed
  76f20c3

  Copy full SHA for 76f20c3

• wrote ConvOp description for conv. neural nets

  gdesjardins

  committed
  79c90fa

  Copy full SHA for 79c90fa

Commits on Jan 19, 2010

• I redone the tests and updated final scores and how long they run

  pascanur

  committed
  61ceb18

  Copy full SHA for 61ceb18

Commits on Jan 17, 2010

• Change 'T.sum' to 'T.mean' in computations of NLL, change the documentation accordingly

  

  Pascal Lamblin

  committed
  3ed2a7f

  Copy full SHA for 3ed2a7f

Commits on Jan 15, 2010

• switching to T.sum when computing negative log likelihood (though T.mean should also be supported) for mlp and logistic_sgd. Second bug fixed was when initializing the weights, we need 6./(n_in+n_o…

  Show description for ff4bd33
  pascanur

  committed
  ff4bd33

  Copy full SHA for ff4bd33

• fixed typo breaking the execution

  

  Yoshua Bengio

  committed
  a8926d3

  Copy full SHA for a8926d3

• minor doc changes + initialization in sqrt(6/(fan_in+fan_out)) instead of sqrt(1/fan_in)

  

  Yoshua Bengio

  committed
  df26db2

  Copy full SHA for df26db2

Commits on Jan 14, 2010

• Removed casting of learning rate to numpy array

  pascanur

  committed
  5c2d4f3

  Copy full SHA for 5c2d4f3

• moved patience improvement 'if-block' inside validation loss improvement 'if-block'

  

  Yoshua Bengio

  committed
  d712ce8

  Copy full SHA for d712ce8

• I changed the pickled file such that it is not divided in minibatches before hand. Note that the code now generates some warning .. this is because of an issue with advanced indexing that Pascal w…

  Show description for 36c0080
  pascanur

  committed
  36c0080

  Copy full SHA for 36c0080

• Several TODO solved

  pascanur

  committed
  2d10ca2

  Copy full SHA for 2d10ca2

Commits on Jan 13, 2010

• merge

  

  James Bergstra

  committed
  882766f

  Copy full SHA for 882766f

• removed categorical_softmax and inserted advanced indexing

  

  James Bergstra

  committed
  092c587

  Copy full SHA for 092c587

Commits on Jan 12, 2010

• fix about the size of the minibatch

  

  Frederic Bastien

  committed
  aea8ac4

  Copy full SHA for aea8ac4

• I changed the code to do a validation test every epoch. While this is not the best option, it was straight forward to implement, and from the discussion in the emails I got the feeling there is no…

  Show description for 4e572e8
  pascanur

  committed
  4e572e8

  Copy full SHA for 4e572e8

Commits on Jan 11, 2010

• Minibatches of size 20

  pascanur

  committed
  2aceb63

  Copy full SHA for 2aceb63

Commits on Jan 8, 2010

• A first draft of mlp tutorial + small changes to the logistic regression

  pascanur

  committed
  840abb3

  Copy full SHA for 840abb3

• Switched to new theano.shared and theano.function

  pascanur

  committed
  fc40c22

  Copy full SHA for fc40c22

• Stats for logistic_sgd and mlp

  pascanur

  committed
  750a2cc

  Copy full SHA for 750a2cc