FeedForwardNNTrainer.h

libcudann

* FeedForwardNNTrainer.h

* Created on: 19/nov/2010

* Author: donati

#ifndef FEEDFORWARDNNTRAINER_H_

#define FEEDFORWARDNNTRAINER_H_

#include "FeedForwardNN.h"

#include "LearningSet.h"

#include "ErrorFunctions.h"

#include "CudaActivationFunctions.cuh"

#include "CudaErrorFunctions.cuh"

#define TRAIN_CPU 0

#define TRAIN_GPU 1

#define ALG_BP 0

#define ALG_BATCH 1

#define SHUFFLE_OFF 0

#define SHUFFLE_ON 1

#define PRINT_ALL 0

#define PRINT_MIN 1

#define PRINT_OFF 2

class FeedForwardNNTrainer {

public:

FeedForwardNNTrainer();

virtual ~FeedForwardNNTrainer();

//choose a net to operate on and save after the training

void selectNet(FeedForwardNN &);

//choose the training set

void selectTrainingSet(LearningSet &);

//choose the test set. if this is set the error rate is computed on test set instead of training set

void selectTestSet(LearningSet &);

//choose a net to save the best network trained so far after each epoch. mse on test set is the criterion

void selectBestMSETestNet(FeedForwardNN &);

//choose a net to save the best network trained so far after each epoch. mse on train set + mse on test set is the criterion

void selectBestMSETrainTestNet(FeedForwardNN &);

//choose a net to save the best network trained so far after each epoch. percentage as classifier is the criterion

void selectBestClassTestNet(FeedForwardNN &);

//starts the training using params. n is the number of parameters

//the first 2 elements of params are where the training will be executed (TRAIN_CPU,TRAIN_GPU)

//and the training algorithm (ALG_BP,ALG_BATCH...). the other parameters are algorithm dependent

//returns the best MSE on test set (or train set if test set isn't specified)

//printtype specifies how much verbose will be the execution (PRINT_ALL,PRINT_MIN,PRINT_OFF)

float train(const int n, const float * params,const int printtype=PRINT_ALL);

private:

//backpropagation training on host

//n is the number of parameters. parameters are (float array):

//desired error, max_epochs, epochs_between_reports, learning_rate, momentum (using momentum is 20% slower), shuffle (SHUFFLE_ON or SHUFFLE_OFF), error function (ERROR_TANH or ERROR_LINEAR)

float trainCpuBp(const int n, const float * params, const int printtype);

//batch training on host

//n is the number of parameters. parameters are (float array):

//desired error, max_epochs, epochs_between_reports, learning_rate, momentum (using momentum is 20% slower), shuffle (SHUFFLE_ON or SHUFFLE_OFF), error function (ERROR_TANH or ERROR_LINEAR)void FeedForwardNNTrainer::trainCpuBatch(const int n, const float * params){

float trainCpuBatch(const int n, const float * params, const int printtype);

//batch training on device

//n is the number of parameters. parameters are (float array):

float trainGPUBatch(const int n, const float * params, const int printtype);

//computes a single instance forward of the backpropagation training

void stepForward(float * values, const float * weights, const int * actFuncts, const int numOfLayers, const int * layersSize, const int numOfInputsPerInstance, const float * trainingSetInputs, const int * offsetIns, const int * offsetWeights, const int * offsetOuts, const int * order, const int instance);

//computes a single instance backward of the backpropagation training

void stepBack(const float * values, const float * weights, float * deltas, const int * actFuncts, const int numOfLayers, const int * layersSize, const int numOfOutputsPerInstance, const float * trainingSetOutputs, const int * offsetWeights, const int * offsetDeltas, const int * offsetOuts, const int * order, const int instance, const int errorFunc);

//update the weights using the deltas

void weightsUpdate(const float * values, const float * weights, float * weightsToUpdate, const float * deltas, const int numOfLayers, const int * layersSize, const int * offsetIns, const int * offsetWeights, const int * offsetDeltas, const float momentum, float * oldWeights, float learningRate);

//GPU computes all the instances forward of the backpropagation training

void GPUForward(float * devValues, const float * devWeights, const int * actFuncts, const int numOfLayers, const int * layersSize, const int numOfInstances, const int * offsetIns, const int * offsetWeights, const int * offsetOuts);

//GPU computes all the instances backward of the backpropagation training

void GPUBack(const float * devValues,const float * devWeights,float * devDeltas,const int * actFuncts,const int numOfLayers,const int *layersSize,const int numOfInstances,const int numOfOutputsPerInstance,const float * devTrainingSetOutputs,const int *offsetWeights,const int *offsetDeltas,const int * offsetOuts, const int errorFunc);

//GPU updates the weights for all the instances

void GPUUpdate(const float * devValues,float * devWeights,const float *devDeltas, const int numOfLayers, const int * layersSize, const int numOfInstances, const int * offsetIns,const int * offsetWeights,const int * offsetDeltas,const float momentum,float * devOldWeights,const float learningRate);

//GPU computes the MSE on a set

float GPUComputeMSE(float * devValues, const float * devWeights, const int * actFuncts, const int numOfLayers, const int * layersSize, const int numOfInstances, const int numOfOutputsPerInstance,const float * devSetOutputs,const int * offsetIns, const int * offsetWeights, const int * offsetOuts);

//GPU computes the classification percentage on a set

float GPUclassificatePerc(float * devValues, const float * devWeights, const int * actFuncts, const int numOfLayers, const int * layersSize, const int numOfInstances, const int numOfOutputsPerInstance,float * devSetOutputs,const int * offsetIns, const int * offsetWeights, const int * offsetOuts);

FeedForwardNN * net;

LearningSet * trainingSet;

LearningSet * testSet;

FeedForwardNN * bestMSETestNet;

FeedForwardNN * bestMSETrainTestNet;

FeedForwardNN * bestClassTestNet;

};

#endif /* FEEDFORWARDNNTRAINER_H_ */