Skip to content

Latest commit

 

History

History
70 lines (69 loc) · 2.53 KB

ROADMAP.md

File metadata and controls

70 lines (69 loc) · 2.53 KB
Raw
  • Cover core with tests
  • Move main.go from root to other destination
  • New layers and learning optimization
    • Softmax layer;
    • Maxout layer;
    • Dropout layer;
    • Optimization for learning;
    • Bias;
  • Add new operations
    • Convolve2D
    • Flatten
    • Pool2D
    • ZeroPadding
    • Determinant
    • ContoursPadding
    • Reshape
  • Refactor code and rethink structure of library
  • Remove some legacy and experemintal code to other branches WIP
  • Error types and wrapping them WIP
  • Test cases for most of math functions WIP
    • ActivationTanh
    • ActivationTanhDerivative
    • ActivationSygmoid
    • ActivationSygmoidDerivative
    • Rot2D90
    • Rot2D180
    • Rot2D270
    • ActivationArcTan
    • ActivationArcTanDerivative
    • ActivationSoftPlus
    • ActivationSoftPlusDerivative
    • ActivationGaussian
    • ActivationGaussianDerivative
    • [x] Add (element-wise) Not needed cause of gonum usage
    • [x] Sub (element-wise) Not needed cause of gonum usage
    • [x] Transpose Not needed cause of gonum usage
    • [x] Multiply Not needed cause of gonum usage
    • [x] HadamardProduct Not needed cause of gonum usage
    • [ ] MSE Not needed cause of gonum usage
    • Convolve2D
    • Flatten
    • Pool2D
    • ZeroPadding
    • Im2Col
    • ContoursPadding
    • Reshape and its Unsafe version
  • Test cases for layers and its methods
    • Convolutional WIP
    • Fully connected WIP
    • ReLU
    • Leaky ReLU
    • Pooling
  • Gonum integration
  • Use of goroutines for boosting calculations
    • [x] ZeroPadding Slow down perfomance
    • [X] Pool2D Slow down perfomance
    • [x] Im2Col Slow down perfomance
    • [x] Flatten Slow down perfomance
    • [ ] ContoursPadding Slow down perfomance
  • Benchmarks. Do we really need it since this is just library for studying purposes? WIP
  • Padding for convolutional layer
  • Write theoretical documents on most of functions (on every would be even better)
  • New struct of examples folder (split it on different types of tasks for neural networks)
  • Consider float32 as extension
  • Improve README's WIP
  • Graphviz pretty print. WIP
  • Add CI on https://travis-ci.com
  • Import/Export activation functions and its derivatives for JSON files.

Updated at: 2020-10-11