·
2 commits
to master
since this release
Introduction :
We are very enthusiastic about this release, we changed all the library to obtain a very performant and simplistically one. We hope that the future will be radiant for this project and that great improvement will come quickly. For the moment the library is functional and rather stable, even if there is still a lot of work as for the implementation of OpenCL.
Changelog :
GENERAL :
- New philosophy : tensor based library.
- Updated the project website.
- Updated the documentation.
- Added a wiki on github.
- Updated the README.md.
CORE :
- Fixed a bug with the error system.
- Deleted String class, use std::string now.
- Deleted File class, use std::fstream now.
- Deleted Thread class, use std::thread.
- Changed the log system to a C++ one.
MLP :
- Deleted the module (replaced by the NeuralNet module).
UTILITY :
- Added a n-tensor system.
- Added a tensor shape system.
- Added an activator function struct (normalized).
- Added an approximation function of derivative.
- Added a gradient struct (old and actual gradient).
- Added support for the Sigmoid activation function.
- Added support to the MNIST database (with a specific loader).
NEURALNET :
- Added a new module : NeuralNet. It will store all neural networks (as a toolkit).
- Added a fully connect layer class.
- Added a pure virtual layer class.
- Added a pure virtual optimizer class (for gradient).
- Added a pure virtual cost class (for cost function).
- Added support to the Fully Connect Layer.
- Added support to the Mean Squared Error cost function.
- Added support to the Stochastic Gradient Descent method.
- Added a Random Initializer.
- Added an Uniform Initializer.
- Added a Zero Initializer.
- Added a convolutional layer class (3D and 2D).
- Added a pooling layer class.
- Added a relu layer class.