Commits

Branch selector

 master

User selector

Wainberg

Datepicker

All time

Commit History

Commits on May 7, 2015

• Don't validate after the first epoch; allow run() to be called multiple times on the same optimizer

  Wainberg

  committed
  c95d266

  Copy full SHA for c95d266

Commits on Apr 19, 2015

• Option to stop SGD from printing

  Wainberg

  committed
  2cd89e7

  Copy full SHA for 2cd89e7

• Box targets if not already boxed

  Wainberg

  committed
  21b6192

  Copy full SHA for 21b6192

Commits on Apr 18, 2015

• Give an error when targets have the wrong shape

  Wainberg

  committed
  2766a77

  Copy full SHA for 2766a77

Commits on Feb 24, 2015

• Re-enable check for finiteness of loss in NeuralNet

  Show description for c0df08c
  Wainberg

  committed
  c0df08c

  Copy full SHA for c0df08c

• Exit gracefully if there's a KeyboardInterrupt before the first validation epoch

  Wainberg

  committed
  aac7915

  Copy full SHA for aac7915

Commits on Feb 21, 2015

• Only import skdata when using MNISTDataProvider

  Wainberg

  committed
  550f2c1

  Copy full SHA for 550f2c1

• Exit on KeyboardInterrupt, once finished cleaning up

  Wainberg

  committed
  7354961

  Copy full SHA for 7354961

• Added DummyProgressMonitor that doesn't report anything

  Wainberg

  committed
  5cc6dd6

  Copy full SHA for 5cc6dd6

• Correction to previous

  Wainberg

  committed
  676c334

  Copy full SHA for 676c334

• Reversal of previous commit - make train error consistently a float, and also L1/L2 penalty

  Wainberg

  committed
  1c38ddd

  Copy full SHA for 1c38ddd

• Linear regression layer returns loss as GPUArray, not float, to be consistent with SoftmaxLayer

  Wainberg

  committed
  97a2f5a

  Copy full SHA for 97a2f5a

Commits on Feb 19, 2015

• Correction to previous dropout changes

  Wainberg

  committed
  6847eae

  Copy full SHA for 6847eae

• Error message when number of inputs does not match number of prev layer's outputs

  Wainberg

  committed
  d9d526b

  Copy full SHA for d9d526b

• More informative error messages in linalg.py

  Wainberg

  committed
  444e200

  Copy full SHA for 444e200

• Correction to previous - manually check contiguity instead

  Wainberg

  committed
  6b6e7b5

  Copy full SHA for 6b6e7b5

• Auto-check CPU arrays for contiguity before copying to GPU, in data_providers.py

  Wainberg

  committed
  0e77210

  Copy full SHA for 0e77210

Commits on Feb 17, 2015

• Fix crash in LinearRegressionLayer with default weight_scale

  Wainberg

  committed
  5943131

  Copy full SHA for 5943131

• Merge branch 'master' of github.com:Wainberg/hebel

  Wainberg

  committed
  e60c9d4

  Copy full SHA for e60c9d4

• Made dropout a probability instead of a boolean

  Show description for 46bc135
  Wainberg

  committed
  46bc135

  Copy full SHA for 46bc135

Commits on Feb 9, 2015

• Fix (?) to CUDA library loading issue

  Wainberg

  committed
  bc428b7

  Copy full SHA for bc428b7

Commits on May 7, 2014

• Fixed block size in kernels; switched to 'square' 16x16x1 blocks

  Wainberg

  committed
  b66da5e

  Copy full SHA for b66da5e

• Windows compatibility fixes

  Show description for d510fa4
  Wainberg

  committed
  d510fa4

  Copy full SHA for d510fa4