NeuralNetwork.h

Go to the documentation of this file.

// n2p2 - A neural network potential package
// Copyright (C) 2018 Andreas Singraber (University of Vienna)
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program.  If not, see <https://www.gnu.org/licenses/>.
 
#ifndef NEURALNETWORK_H
#define NEURALNETWORK_H
 
#include <fstream>   // std::ofstream
#include <string>    // std::string
#include <vector>    // std::vector
 
namespace nnp
{
 
class NeuralNetwork
{
public:
    enum ActivationFunction
    {
        AF_UNSET,
        AF_IDENTITY,
        AF_TANH,
        AF_LOGISTIC,
        AF_SOFTPLUS,
        AF_RELU,
        AF_GAUSSIAN,
        AF_COS,
        AF_REVLOGISTIC,
        AF_EXP,
        AF_HARMONIC
    };
 
    enum ModificationScheme
    {
        MS_ZEROBIAS,
        MS_ZEROOUTPUTWEIGHTS,
        MS_FANIN,
        MS_GLOROTBENGIO,
        MS_NGUYENWIDROW,
        MS_PRECONDITIONOUTPUT
    };
 
    NeuralNetwork(
                int                              numLayers,
                int const* const&                numNeuronsPerLayer,
                ActivationFunction const* const& activationFunctionsPerLayer);
    ~NeuralNetwork();
    // Prevent copying.
    //NeuralNetwork(const NeuralNetwork&) = delete;
    //NeuralNetwork(NeuralNetwork&&) = delete;
    void                     setNormalizeNeurons(bool normalizeNeurons);
    int                      getNumNeurons() const;
    int                      getNumConnections() const;
    int                      getNumWeights() const;
    int                      getNumBiases() const;
    void                     setConnections(double const* const& connections);
    void                     getConnections(double* connections) const;
    void                     initializeConnectionsRandomUniform(
                                                            unsigned int seed);
    void                     modifyConnections(
                                        ModificationScheme modificationScheme);
    void                     modifyConnections(
                                         ModificationScheme modificationScheme,
                                         double             parameter1,
                                         double             parameter2);
    void                     setInput(double const* const& input) const;
    void                     setInput(std::size_t const index,
                                      double const      value) const;
    void                     getOutput(double* output) const;
    void                     propagate();
    void                     calculateDEdG(double* dEdG) const;
    void                     calculateDEdc(double* dEdc) const;
    void                     calculateDFdc(double*              dFdc,
                                           double const* const& dGdxyz) const;
    void                     writeConnections(std::ofstream& file) const;
    void                     getNeuronStatistics(long*   count,
                                                 double* min,
                                                 double* max,
                                                 double* sum,
                                                 double* sum2) const;
    void                     resetNeuronStatistics();
    //void   writeStatus(int, int);
    long                     getMemoryUsage();
    std::vector<std::string> info() const;
 
private:
    typedef struct
    {
        long    count;
        double  x;
        double  value;
        double  dfdx;
        double  d2fdx2;
        double  bias;
        double  dxdG;
        double  min;
        double  max;
        double  sum;
        double  sum2;
        double* weights;
    } Neuron;
 
    typedef struct
    {
        int                numNeurons;
        int                numNeuronsPrevLayer;
        ActivationFunction activationFunction;
        Neuron*            neurons;
    } Layer;
 
    bool   normalizeNeurons;
    int    numWeights;
    int    numBiases;
    int    numConnections;
    int    numLayers;
    int    numHiddenLayers;
    int*   weightOffset;
    int*   biasOffset;
    int*   biasOnlyOffset;
    Layer* inputLayer;
    Layer* outputLayer;
    Layer* layers;
 
    void   calculateDEdb(double* dEdb) const;
    void   calculateDxdG(int index) const;
    void   calculateD2EdGdc(int                  index,
                            double const* const& dEdb,
                            double*              d2EdGdc) const;
    void   allocateLayer(Layer&             layer,
                         int                numNeuronsPrevLayer,
                         int                numNeurons,
                         ActivationFunction activationFunction);
    void   propagateLayer(Layer& layer, Layer& layerPrev);
};
 
NeuralNetwork::ActivationFunction activationFromString(std::string c);
 
}
 
#endif