Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Oct 16, 2017

• Remove relics from old codebase

  larspars

  authored
  011d5b1

  Copy full SHA for 011d5b1

• Remove unecessary printing during initial read of data

  larspars

  authored
  fdc7ae6

  Copy full SHA for fdc7ae6

• Update readme

  larspars

  authored
  eefbeb9

  Copy full SHA for eefbeb9

• Update readme

  larspars

  authored
  4abfb89

  Copy full SHA for 4abfb89

Commits on Sep 18, 2017

• Cleanup BELU code

  larspars

  authored
  c0b53f2

  Copy full SHA for c0b53f2

Commits on Sep 15, 2017

• Add link to the Bipolar Activation Functions paper

  larspars

  authored
  df7fba1

  Copy full SHA for df7fba1

Commits on Sep 12, 2017

• Small readme change

  larspars

  authored
  b8e1ebd

  Copy full SHA for b8e1ebd

• Bipolar activations, LSUV init, stochastic skip, zoneout, and more

  larspars

  committed
  4434e53

  Copy full SHA for 4434e53

Commits on Feb 28, 2016

• Updated readme

  larspars

  committed
  9595045

  Copy full SHA for 9595045

• Removed remnants from experiments

  larspars

  committed
  41747de

  Copy full SHA for 41747de

• Removed remnants of fbcunn, moved initialization to LSTM layer

  larspars

  committed
  bc0b8d6

  Copy full SHA for bc0b8d6

Commits on Feb 13, 2016

• Mentioned lrexlib dependency

  larspars

  committed
  38f366a

  Copy full SHA for 38f366a

• Merged in from char-rnn

  larspars

  committed
  38160af

  Copy full SHA for 38160af

• Adding dropout on recurrent connections, IRNN, plus a bunch of experimental code

  larspars

  committed
  179100b

  Copy full SHA for 179100b

Commits on Nov 25, 2015

• Merge pull request #132 from udibr/master

  Show description for a8c1582
  karpathy

  committed
  a8c1582

  Copy full SHA for a8c1582

Commits on Nov 24, 2015

• fix message

  udibr

  committed
  f7a8e74

  Copy full SHA for f7a8e74

• bug fix

  udibr

  committed
  0d55784

  Copy full SHA for 0d55784

Commits on Oct 26, 2015

• Merge pull request #125 from skwerlman/master

  Show description for 106a75c
  karpathy

  committed
  106a75c

  Copy full SHA for 106a75c

• Fix unclear errors

  Show description for b3c7e3c
  skwerlman

  committed
  b3c7e3c

  Copy full SHA for b3c7e3c

Commits on Oct 23, 2015

• Added word embedding support to GRU and RNN networks

  larspars

  committed
  7a9bbc6

  Copy full SHA for 7a9bbc6

• Merged in latest from char-rnn

  larspars

  committed
  bd35304

  Copy full SHA for bd35304

Commits on Oct 19, 2015

• Merge pull request #1 from MakeMeASandwich/patch-1

  Show description for 6acd787
  larspars

  committed
  6acd787

  Copy full SHA for 6acd787

Commits on Oct 18, 2015

• mention fbcunn dependency in readme

  MakeMeASandwich

  committed
  e7afc92

  Copy full SHA for e7afc92

Commits on Oct 16, 2015

• Readme tweaks and clarifications

  karpathy

  committed
  54b4c66

  Copy full SHA for 54b4c66

Commits on Oct 15, 2015

• Added link to char-rnn

  larspars

  committed
  4ef5e4f

  Copy full SHA for 4ef5e4f

• Updated readme

  larspars

  committed
  eb44892

  Copy full SHA for eb44892

• Cleaned up, added explanation

  larspars

  committed
  502c117

  Copy full SHA for 502c117

• Can now backprop into wordvectors

  larspars

  committed
  a8dac34

  Copy full SHA for a8dac34

• More memory efficient word iteration for huge datasets

  larspars

  committed
  83df63d

  Copy full SHA for 83df63d

• Refactored how embeddings are used, and made the choice of optimzer a command line option

  larspars

  committed
  86eee7f

  Copy full SHA for 86eee7f

• Added functionality for continously reading new seed text from stdin

  larspars

  committed
  945f1e6

  Copy full SHA for 945f1e6

Commits on Oct 11, 2015

• speeding up the code by about 15% with faster indexing. I did not imagine this would have such a large effect. Thank you @vseledkin for pointing this out

  karpathy

  committed
  82baeea

  Copy full SHA for 82baeea

Commits on Sep 21, 2015

• adding clarification justifying not normalizing gradient by seqlength

  karpathy

  committed
  4297a9b

  Copy full SHA for 4297a9b

Commits on Sep 20, 2015

• changing default LSTM initialization to use biases of 1.0 for the forget gates, which is a common trick and encourages remembering in the beginning of training. I ran 2 medium-sized experiments and…

  Show description for 0dfeaa4
  karpathy

  committed
  0dfeaa4

  Copy full SHA for 0dfeaa4

Commits on Aug 27, 2015

• Merge pull request #92 from ajkl/patch-1

  Show description for 065c5ff
  karpathy

  committed
  065c5ff

  Copy full SHA for 065c5ff

Pagination

Previous

Next