sdadia / deep-learning-and-optimization

Dependencies

• Xtensor : https://github.com/QuantStack/xtensor
• BLAS & LAPACK
• Gtest : https://github.com/google/googletest
• Cmake
• Make
• Doxygen
• Git

Install Dependencies

sudo apt-get update ; sudo apt-get upgrade

• BLAS and LAPACK, Make, libstdc++, GCC, Doxygen, Gtest:

sudo apt-get install build-essential git libblas-dev liblapack-dev libgtest-dev doxygen

• Xtensor : https://github.com/QuantStack/xtensor

cd $HOME ; git clone https://github.com/QuantStack/xtensor
cd xtensor; mkdir build && cd build;
cmake -DBUILD_TESTS=ON -DDOWNLOAD_GTEST=ON ..
make
make xtest
sudo make install

Compilation & Install

This will create static library in build/lib & executibles in build/bin/

make all -j4 -s -O      # to make static library and test executible
make docs               # to create documentation

make clean # to remove all the above

Create new layer

Just create a new .hpp and .cpp file with the corresponding layer name

• implement setUp(), forward(), backward() functions

sigmoid.hpp

#include "common.hpp"

namespace dpl
{
    class MySigmoidLayer
    {
        void setUp(std::vector<ftensor_t> inputs);

        void forward();

        void backward(std::vector<ftensor_t> grad_from_top);
    }
}

sigmoid.cpp

#include "common.hpp"
#include "sigmoid.hpp"

namespace dpl
{
    void MySigmoidLayer::setUp(std::vector<ftensor_t> inputs)
    {
        // code here
    }

    void MySigmoidLayer::forward()
    {
        // forward prop code here
    }   

    void MySigmoidLayer::backward(std::vector<ftensor_t> grad_from_top)
    {
        // back prop code here
    }
    
}