Skip to content

Version 0.3
Compare
Choose a tag to compare
@mroodschild mroodschild released this 24 Apr 23:30
· 177 commits to master since this release
0.3
cd22e1d

24/04/2018

====== Version 0.3 =======

Performance Improvements
time execution reduced by 75%

Internal Architecture Vectorized

EJML Updated 0.30 -> 0.34

Neural Network

  • removed outputAll

Feedforward

  • outputLayers -> activations
  • activationsDropout
  • fixed getParameters

Layers

  • outputDropout

Initializations

  • HeInit

LossFunctions

  • Improvement CrossEntropy
  • Added Logistic Loss Function
  • MSE vectorized
  • RMSE vectorized

Backpropagation

  • removed gradient
  • removed minibatch
  • fixed L2 regularization
  • Performance incresed

Training Algorithms

  • Conjugate Gradient Added
  • Scaled Conjugate Gradient Added
  • Scaled Conjugate Gradient with minibatch added
  • Clean SGD
  • SGD Numeric Added
  • TrainingALgorithm Reorganized

Beta Rules Added for Conjugate Gradient

  • Fletcher Reeves
  • PolakRebiere

Gradients

  • DropoutGradient
  • NoiseGradient
  • NumericGradient
  • StandardGradient
  • Gradient
  • GradientFactory

Update Rules

  • Momentum
  • SGC
  • Update
  • UpdateRule

TransferFunctions

  • update EJML 0.30 -> 0.34

Test Added

Assets 5
Loading