Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit history

Commits on Jun 5, 2014

• added render weights by layer to configuratio

  

  agibsonccc

  committed
  1d39d7f

  Copy full SHA for 1d39d7f

Commits on Jun 1, 2014

• fix filter render for output layer and generalize better

  

  agibsonccc

  committed
  5b2acb6

  Copy full SHA for 5b2acb6

Commits on May 30, 2014

• added ability to customize deep autoencoder objective function on output

  

  agibsonccc

  committed
  ae3d27c

  Copy full SHA for ae3d27c

• added objects for rendering reconstructions from neural networks,multi layer and deep autoencoders

  

  agibsonccc

  committed
  ec43d0f

  Copy full SHA for ec43d0f

Commits on May 29, 2014

• working deep autoencoder demo, very simple 2 label approach

  

  agibsonccc

  committed
  381c1e0

  Copy full SHA for 381c1e0

• working deep autoencoder demo, very simple 2 label approach

  

  agibsonccc

  committed
  73cac9e

  Copy full SHA for 73cac9e

• squared loss should not be negative on the neural net itself

  

  agibsonccc

  committed
  6b8c159

  Copy full SHA for 6b8c159

• added loss function by layer

  

  agibsonccc

  committed
  6cf7f11

  Copy full SHA for 6cf7f11

• improved deep autoencoder example

  

  agibsonccc

  committed
  4c8e2bb

  Copy full SHA for 4c8e2bb

• improved deep autoencoder example

  

  agibsonccc

  committed
  95cce14

  Copy full SHA for 95cce14

Commits on May 28, 2014

• added a few getters and setters for deep autoencoder, allows for code layer specific overrides to deal with certain problems such as document retrieval or images

  

  agibsonccc

  committed
  2c84bc2

  Copy full SHA for 2c84bc2

Commits on May 27, 2014

• added base state tracker

  

  agibsonccc

  committed
  690b176

  Copy full SHA for 690b176

Commits on May 26, 2014

• added distributed deep autoencoders

  

  agibsonccc

  committed
  33c3afd

  Copy full SHA for 33c3afd

Commits on May 25, 2014

• added some more functionality to word2vec and also effectively halved the memory footprint changing the representation to floats rather than matrices, still playing with what to do with the neural …

  Show description for 975effa
  

  agibsonccc

  committed
  975effa

  Copy full SHA for 975effa

• minor changes/added loading google binary models

  

  agibsonccc

  committed
  f11deb5

  Copy full SHA for f11deb5

• updated readme

  

  agibsonccc

  committed
  4d6c47c

  Copy full SHA for 4d6c47c

• clean up worker a bit

  

  agibsonccc

  committed
  76ca250

  Copy full SHA for 76ca250

Commits on May 21, 2014

• newer pipeline; fixed end to end

  

  agibsonccc

  committed
  6437ef5

  Copy full SHA for 6437ef5

Commits on May 19, 2014

• changed deep autoencoder example

  

  agibsonccc

  committed
  cc3e8d4

  Copy full SHA for cc3e8d4

Commits on May 18, 2014

• working deep autoencoders test

  

  agibsonccc

  committed
  7577441

  Copy full SHA for 7577441

Commits on May 17, 2014

• added gaussian hidden units for RBMs

  

  agibsonccc

  committed
  f1e9d6a

  Copy full SHA for f1e9d6a

• last minute fixes to deep autoencoder

  

  agibsonccc

  committed
  0259702

  Copy full SHA for 0259702

• deep autoencoders wrt images work now, still playing around with different kinds of inputs yet

  

  agibsonccc

  committed
  35ec02a

  Copy full SHA for 35ec02a

Commits on May 15, 2014

• fixed hidden layer initialization

  

  agibsonccc

  committed
  75c395b

  Copy full SHA for 75c395b

• shorten test times, add bintray distribution

  

  agibsonccc

  committed
  b474000

  Copy full SHA for b474000

• centralize initialization of networks from configuration in configuration

  

  agibsonccc

  committed
  82d5438

  Copy full SHA for 82d5438

Commits on May 14, 2014

• dont render weights on RBM test prepare for release on bintray

  

  agibsonccc

  committed
  eb4b054

  Copy full SHA for eb4b054

• fixed test; thanks!

  

  agibsonccc

  committed
  810cf1d

  Copy full SHA for 810cf1d

• stop at flip of sign with backprop when it gets below zero

  

  agibsonccc

  committed
  c67255f

  Copy full SHA for c67255f

Commits on May 13, 2014

• ensure decoder is initialized on finetune

  

  agibsonccc

  committed
  8b30c65

  Copy full SHA for 8b30c65

• changed the first visible unit for the decoder to be gaussian, also added ability to configure the visible and hidden unit by layer falling back to a specified default otherwise

  

  agibsonccc

  committed
  2211a72

  Copy full SHA for 2211a72

• add initialization check for when a hidden layer hasnt been pretrained yet

  

  agibsonccc

  committed
  6bf35ff

  Copy full SHA for 6bf35ff

• added model saving persistence checks

  

  agibsonccc

  committed
  20526ff

  Copy full SHA for 20526ff

Commits on May 12, 2014

• added tensor and fourdtensor adagrad

  

  agibsonccc

  committed
  f0c2cd4

  Copy full SHA for f0c2cd4

• fixed issue with multi threaded networks and the best loss objective on finetune

  

  agibsonccc

  committed
  ed1e585

  Copy full SHA for ed1e585