Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit history

Commits on Aug 24, 2014

• Update LFWRBMExample.java

  Show description for 8e42942
  chrisvnicholson

  committed
  8e42942

  Copy full SHA for 8e42942

Commits on Aug 16, 2014

• fix hidden/visible unit init

  

  agibsonccc

  committed
  f9e2b0e

  Copy full SHA for f9e2b0e

Commits on Aug 7, 2014

• add uniform way of doing fast element wise operations

  

  agibsonccc

  committed
  6b4429d

  Copy full SHA for 6b4429d

Commits on Aug 2, 2014

• little workaround for complex ndarray columns

  

  agibsonccc

  committed
  08cd051

  Copy full SHA for 08cd051

Commits on Jul 21, 2014

• added proper dimension wise iteration (think for ufuncs in numpy

  

  agibsonccc

  committed
  83aa782

  Copy full SHA for 83aa782

Commits on Jul 10, 2014

• added small tweet example

  Show description for 91d8c80
  

  agibsonccc

  committed
  91d8c80

  Copy full SHA for 91d8c80

• added more tweet examples

  Show description for ca5d6e9
  

  agibsonccc

  committed
  ca5d6e9

  Copy full SHA for ca5d6e9

• revert softmax change, probability distribution seems to have smooted out on RNTNs, making note here: may need to work out row maxes in softmax function as opposed to columns, may have been cause o…

  Show description for 42967e8
  

  agibsonccc

  committed
  42967e8

  Copy full SHA for 42967e8

Commits on Jul 9, 2014

• better softmax function

  

  agibsonccc

  committed
  fe41ce2

  Copy full SHA for fe41ce2

• fix derivative on tanh

  

  agibsonccc

  committed
  90c0641

  Copy full SHA for 90c0641

Commits on Jul 8, 2014

• fixed some matrix multiplication issues

  

  agibsonccc

  committed
  afa58b9

  Copy full SHA for afa58b9

Commits on Jul 7, 2014

• more robust binarization algorithm; the previous one fell down with odd things like tweets

  

  agibsonccc

  committed
  a8ee4c4

  Copy full SHA for a8ee4c4

Commits on Jul 6, 2014

• minor tree binarizer fixes, also ported RNTN to float matrices, added float matrix adagrad, and added activations for float matrices (expect memory reduction by a metric ton here soon

  

  agibsonccc

  committed
  e54d511

  Copy full SHA for e54d511

Commits on Jul 5, 2014

• add basic example for rntn training and a few predict methods

  

  agibsonccc

  committed
  87516f2

  Copy full SHA for 87516f2

• added setting gold labels for trees for training as well as supervised learning pipeline for RNTNs

  

  agibsonccc

  committed
  5817e73

  Copy full SHA for 5817e73

Commits on Jul 2, 2014

• fixed error from latest commit on the parser example

  

  agibsonccc

  committed
  21201d8

  Copy full SHA for 21201d8

Commits on Jul 1, 2014

• Update DeepAutoEncoderSDA.java

  chrisvnicholson

  committed
  315d40c

  Copy full SHA for 315d40c

• Update MovingWindowExample.java

  chrisvnicholson

  committed
  74baa70

  Copy full SHA for 74baa70

• Update StackedDenoisingAutoEncoderExample.java

  chrisvnicholson

  committed
  48f8cc3

  Copy full SHA for 48f8cc3

Commits on Jun 30, 2014

• added constituency parser, still need to work on pipeline for iterating through the tree and binding these to be trees with word vectors

  

  agibsonccc

  committed
  6576c86

  Copy full SHA for 6576c86

• added thread safe stemmer

  

  agibsonccc

  committed
  fa5db02

  Copy full SHA for fa5db02

• SHF produces reasonable results on a small dataset; still need to test at scale

  

  agibsonccc

  committed
  63d5e83

  Copy full SHA for 63d5e83

Commits on Jun 29, 2014

• published curves example

  

  agibsonccc

  committed
  60e177b

  Copy full SHA for 60e177b

• bias gradient should always be mean

  

  agibsonccc

  committed
  cf0e334

  Copy full SHA for cf0e334

• SHF somewhat more coherent, still working on some tests yet

  

  agibsonccc

  committed
  78ff6f6

  Copy full SHA for 78ff6f6

Commits on Jun 26, 2014

• remove instrumentation module; aspects not used

  

  agibsonccc

  committed
  d4d71ea

  Copy full SHA for d4d71ea

Commits on Jun 24, 2014

• slightly saner math, missed operators

  

  agibsonccc

  committed
  18fd348

  Copy full SHA for 18fd348

• commiting snapshot, not quite there yet

  

  agibsonccc

  committed
  a7df54b

  Copy full SHA for a7df54b

Commits on Jun 23, 2014

• added more hessian free; packing/unpacking nets, leveraging of the v matrix)

  

  agibsonccc

  committed
  43ae65a

  Copy full SHA for 43ae65a

Commits on Jun 22, 2014

• added progress report to multi epoch datasetiterator

  

  agibsonccc

  committed
  fedda09

  Copy full SHA for fedda09

• norm2 on gradient constraint added

  

  agibsonccc

  committed
  709de20

  Copy full SHA for 709de20

• added gradient constraint to unit norm

  

  agibsonccc

  committed
  d9b3dfc

  Copy full SHA for d9b3dfc

• backprop changes for less memory usage

  

  agibsonccc

  committed
  1ab35eb

  Copy full SHA for 1ab35eb

Commits on Jun 19, 2014

• added R operator feed forward (this is for hessian free), output layer dropout

  

  agibsonccc

  committed
  411de5e

  Copy full SHA for 411de5e

Commits on Jun 18, 2014

• fixed merge

  

  agibsonccc

  committed
  d28f81f

  Copy full SHA for d28f81f