Sigmoid.hpp

Go to the documentation of this file.

#ifndef SRC_MLNN_SIGMOID_HPP_
#define SRC_MLNN_SIGMOID_HPP_

#include <mlnn/layer/Layer.hpp>

namespace mic {
namespace mlnn {
namespace activation_function {

template <typename eT=float>
class Sigmoid : public mic::mlnn::Layer<eT> {
public:
    Sigmoid(size_t size_,std::string name_ = "Sigmoid") :
        Sigmoid(size_, 1, 1, name_)
    {

    }


    Sigmoid(size_t height_, size_t width_, size_t depth_,
            std::string name_ = "Sigmoid") :
        Layer<eT>::Layer(height_, width_, depth_,
                height_, width_, depth_,
                LayerTypes::Sigmoid, name_)
    {

    }

    virtual ~Sigmoid() {};

    void forward(bool test = false) {
        // Access the data of both matrices.
        eT* x = s['x']->data();
        eT* y = s['y']->data();

        for (size_t i = 0; i < (size_t)s['x']->rows() * s['x']->cols(); i++) {
            y[i] = 1.0f / (1.0f +::exp(-x[i])); //: float -> expf
        }//: for
    }

    void backward() {
        // Access the data of matrices.
        eT* gx = g['x']->data();
        eT* gy = g['y']->data();
        eT* y = s['y']->data();

        for (size_t i = 0; i < (size_t)g['x']->rows() * g['x']->cols(); i++) {
            // "Pass" the gradient multiplied by the sigmoid derivative.
            gx[i] = gy[i]* (y[i] * (1.0 - y[i]));
        }//: for
    }

    virtual void update(eT alpha_, eT decay_  = 0.0f) { };

protected:
    // Unhiding the template inherited fields via "using" statement.
    using Layer<eT>::g;
    using Layer<eT>::s;

private:
    // Friend class - required for using boost serialization.
    template<typename tmp> friend class mic::mlnn::MultiLayerNeuralNetwork;

    Sigmoid<eT>() : Layer<eT> () { }


};

} /* activation_function */
} /* namespace mlnn */
} /* namespace mic */

#endif /* SRC_MLNN_SIGMOID_HPP_ */