MyMatrix MyQuat::convertToRotationMatrix(void) const
{
MyMatrix result;
GLfloat rotMatrix[16];
float xx = this->w * this->w;
float xy = this->w * this->v.x;
float xz = this->w * this->v.y;
float xw = this->w * this->v.z;
float yy = this->v.x * this->v.x;
float yz = this->v.x * this->v.y;
float yw = this->v.x * this->v.z;
float zz = this->v.y * this->v.y;
float zw = this->v.y * this->v.z;
rotMatrix[0] = 1 - 2 * (yy + zz);
rotMatrix[1] = 2 * (xy - zw);
rotMatrix[2] = 2 * (xz + yw);
rotMatrix[4] = 2 * (xy + zw);
rotMatrix[5] = 1 - 2 * (xx + zz);
rotMatrix[6] = 2 * (yz - xw);
rotMatrix[8] = 2 * (xz - yw);
rotMatrix[9] = 2 * (yz + xw);
rotMatrix[10] = 1 - 2 * (xx + yy);
rotMatrix[3] = rotMatrix[7] = rotMatrix[11] = rotMatrix[12] = rotMatrix[13] = rotMatrix[14] = 0;
rotMatrix[15] = 1;
result.setMyMatrix(rotMatrix);
return result;
}
void * Chi2Lib::computeDifferenceThread( void* ptr){
PartitionDiff *part = (PartitionDiff*) ptr;
MyMatrix<double> *img = part->img;
MyMatrix<double> *grid_x = part->grid_x;
MyMatrix<double> *grid_y = part->grid_y;
double d = part->d;
double w = part->w;
MyMatrix<double> *diffout = part->diffout;
MyLogger::log()->debug("[Chi2Lib][computeDifferenceThread] Computing Chi2 Difference");
double chi2err = 0.0;
for(unsigned int x=part->x1; x < part->x2; ++x){
for(unsigned int y=part->y1; y < part->y2; ++y){
double x2y2 = sqrt(1.0*grid_x->getValue(x,y)*grid_x->getValue(x,y) + 1.0*grid_y->getValue(x,y)*grid_y->getValue(x,y));
double temp = ((1.0-tanh( (x2y2-d/2.0)/w )) - 2.0*img->getValue(x,y))/2.0;
diffout->at(x,y) = temp;
chi2err += temp*temp;
}
}
MyLogger::log()->debug("[Chi2Lib][computeDifferenceThread] Chi2 Error: %f", chi2err);
part->err = chi2err;
return 0;
}
CellMatrix::CellMatrix(const MyMatrix& data): Cells(data.rows()),
Rows(data.rows()), Columns(data.columns())
{
for (unsigned long i=0; i < data.rows(); ++i)
for (unsigned long j=0; j < data.columns(); ++j)
Cells[i].push_back(CellValue(Element(data,i,j)));
}
void MenuSlider::Draw(MyMatrix* pMatProj, MyMatrix* pMatView)
{
float centerx = m_PosX;
float top = m_PosY;
// TODO: take more justifications and vertical/horizontal into account for top and left.
// should ideally be done once instead of every frame.
//if( m_Justification & Justify_Left )
//{
// centerx += m_BarThickness/2;
//}
float emptypos = (m_PosY - m_VisualRange);
MyAssert( m_pSprite );
//m_pSprite = g_pGame->m_pResources->m_pSprites[SL_WhiteSquareResizable];
if( m_pSprite )
{
BufferManager* pBufferManager = m_pGameCore->GetManagers()->GetBufferManager();
m_pSprite->Create( pBufferManager, "MenuSlider", m_BarThickness, m_VisualRange, 0, 1, 0, 1, Justify_CenterX|Justify_Top );
MyMatrix world;
world.SetIdentity();
world.SetTranslation( centerx, top, 0 );
//FIX m_pSprite->SetTint( m_Colors[MSCT_BarColor] );
m_pSprite->Draw( pMatProj, pMatView, &world ); //&g_pGame->m_OrthoMatrix );
m_pSprite->Create( pBufferManager, "MenuSlider", m_HandleWidth, m_HandleHeight, 0, 1, 0, 1, Justify_CenterX|Justify_Top );
world.SetTranslation( centerx, emptypos + m_ValuePerc*m_VisualRange, 0 );
//FIX m_pSprite->SetTint( m_Colors[MSCT_HandleColor] );
m_pSprite->Draw( pMatProj, pMatView, &world ); //&g_pGame->m_OrthoMatrix );
}
}
xlw::CellMatrix::CellMatrix(const MyMatrix& data): Cells(data.size1()),
Rows(data.size1()), Columns(data.size2())
{
for (size_t i=0; i < data.size1(); ++i)
for (size_t j=0; j < data.size2(); ++j)
Cells[i].push_back(CellValue(Element(data,i,j)));
}
MyMatrix MyMatrix::operator+(const MyMatrix& A)
{
MyMatrix result;
result.setN(fN);
result.setM(fM);
result.fMatrix = fMatrix+A.fMatrix;
return result;
}
double BSDeltaWithParamsFD(const MyMatrix& parametersMatrix, double epsilon) {
if (parametersMatrix.columns() != 6 && parametersMatrix.rows() != 1 ) {
throw("Input matrix should be 1 x 5");}
double Spot = parametersMatrix(0,0);
double Strike = parametersMatrix(0,1);
double r = parametersMatrix(0,2);
double d = parametersMatrix(0,3);
double vol = parametersMatrix(0,4);
double expiry = parametersMatrix(0,5);
return BlackScholesDeltaFD(Spot, Strike,r,d,vol,expiry, epsilon);
}
double BSZeroCouponBondWithParams(const MyMatrix& parametersMatrix){
if (parametersMatrix.columns() != 6 && parametersMatrix.rows() != 1 ) {
throw("Input matrix should be 1 x 5");}
double Spot = parametersMatrix(0,0);
double Strike = parametersMatrix(0,1);
double r = parametersMatrix(0,2);
double d = parametersMatrix(0,3);
double vol = parametersMatrix(0,4);
double expiry = parametersMatrix(0,5);
return BlackScholesZeroCouponBond(Spot, Strike,r,d,vol,expiry);
}
CellMatrix::CellMatrix(const MyMatrix& data):pimpl(new defaultCellMatrixImpl(data.rows(),data.columns()))
{
for(size_t i(0); i < data.rows(); ++i)
{
for(size_t j(0); j < data.columns(); ++j)
{
(*pimpl)(i,j) = data(i,j);
}
}
}
void MyMatrix::RotateZ(float angle)
{
float sinAngle, cosAngle;
sinAngle = sinf ( angle * PI / 180.0f );
cosAngle = cosf ( angle * PI / 180.0f );
MyMatrix temp;
temp.LoadIndentity();
temp.m[0][0] = temp.m[1][1] = cosAngle;
temp.m[0][1] = sinAngle;
temp.m[1][0] = -sinAngle;
MatrixMultiply(&temp,this);
}
void operator/(const MyMatrix &b){ // transpose
for(int i = 0; i < _dim; i++){
for(int j = 0; j < _dim; j++){
vecvec[i][j] = b.get(j,i);
}
}
}
void operator-=(const MyMatrix &b){
for(int i = 0; i < _dim; i++){
for(int j = 0; j < _dim; j++){
int tmp= (int) (vecvec[i][j]- b.get(i, j));
vecvec[i][j]=tmp;
}
}
}
CellMatrix //
BSGreeksFD(const MyMatrix& parametersMatrix,double epsilon) {
if (parametersMatrix.columns() != 6 && parametersMatrix.rows() != 1 ) {
throw("Input matrix should be 1 x 5");}
double Spot = parametersMatrix(0,0);
double Strike = parametersMatrix(0,1);
double r = parametersMatrix(0,2);
double d = parametersMatrix(0,3);
double vol = parametersMatrix(0,4);
double expiry = parametersMatrix(0,5);
CellMatrix resultMatrix(1,5);
resultMatrix(0,0) = BlackScholesDeltaFD(Spot,Strike,r,d,vol,expiry,epsilon);
resultMatrix(0,1) = BlackScholesGammaFD(Spot,Strike,r,d,vol,expiry,epsilon);
resultMatrix(0,2) = BlackScholesVegaFD(Spot,Strike,r,d,vol,expiry,epsilon);
resultMatrix(0,3) = BlackScholesRhoFD(Spot,Strike,r,d,vol,expiry,epsilon);
resultMatrix(0,4) = BlackScholesThetaFD(Spot,Strike,r,d,vol,expiry,epsilon);
return resultMatrix;
}
CellMatrix // returns delta, gamma, vega, rho, theta
BSGreeks(const MyMatrix& parametersMatrix) {
if (parametersMatrix.columns() != 6 && parametersMatrix.rows() != 1 ) {
throw("Input matrix should be 1 x 5");}
double Spot = parametersMatrix(0,0);
double Strike = parametersMatrix(0,1);
double r = parametersMatrix(0,2);
double d = parametersMatrix(0,3);
double vol = parametersMatrix(0,4);
double expiry = parametersMatrix(0,5);
CellMatrix resultMatrix(1,5);
resultMatrix(0,0) = BlackScholesDelta(Spot,Strike,r,d,vol,expiry);
resultMatrix(0,1) = BlackScholesGamma(Spot,Strike,r,d,vol,expiry);
resultMatrix(0,2) = BlackScholesVega(Spot,Strike,r,d,vol,expiry);
resultMatrix(0,3) = BlackScholesRho(Spot,Strike,r,d,vol,expiry);
resultMatrix(0,4) = BlackScholesTheta(Spot,Strike,r,d,vol,expiry);
return resultMatrix;
}
/**********************************************************
*
* SetViewTransformation
*
* parameters IN :
* MyMatrix & viewTransformation
*
*********************************************************/
void ViewPointRec::SetViewTransformation (MyMatrix & viewTransformation) {
viewTransformation.Identity();
viewTransformation[0][0] = -sinTheta;
viewTransformation[0][1] = -cosTheta * cosPhi;
viewTransformation[0][2] = -cosTheta * sinPhi;
viewTransformation[1][0] = cosTheta;
viewTransformation[1][1] = -sinTheta * cosPhi;
viewTransformation[1][2] = -sinTheta * sinPhi;
viewTransformation[2][1] = sinPhi;
viewTransformation[2][2] = -cosPhi;
viewTransformation[3][2] = rho;
}
MyVector operator*(const MyMatrix& lhs ,const MyVector& rhs) {
MyVector y = MyVector(lhs.rows_);
if (lhs.columns_ == rhs.size()) {
for(long long int i = 1; i <= lhs.rows_; i++) {
for(long long int j = 1; j <= lhs.columns_; j++) {
y[i-1] += rhs[j-1] *lhs.at(i,j);
}
}
} else {
cout << "Matrix-Vektor-Multiplikation nicht möglich Dimensionen stimmen nicht überein!" << endl;
}
return y;
}
MyMatrix MyMatrix::operator*(const MyMatrix& rhs) {
MyMatrix lhs = *this;
auto n=lhs.rows_;
auto m = lhs.columns_;
auto p = rhs.columns_;
MyMatrix y = MyMatrix(n,p);
if (m == rhs.rows_) {
for(long long int i = 1; i <= n; i++) {
for(long long int j = 1; j <= p; j++) {
y.mval_[(i-1)*(p) + (j-1)]=0;
for(long long int k = 1; k <= m; k++) {
y.mval_[(i-1)*(p) + (j-1)] += lhs.at(i,k)*rhs.at(k,j) ;
}
}
}
}
else {
cout << "Dimensionen stimmen nicht überein!" << endl;
}
return y;
}
void operator*=(const MyMatrix &b){
for(int i = 0; i < _dim; i++){
std::vector<float> temporalVectorA((unsigned long) _dim,0);
for (int h = 0; h < _dim; h++){
temporalVectorA[h] = vecvec[i][h];
}
for(int k = 0; k < _dim; k++){
std::vector<float> temporalVectorB((unsigned long) _dim,0);
for(int l = 0; l < _dim; l++){
temporalVectorB[l] = b.get(l, k);
}
vecvec[i][k] = vectorMultiply(temporalVectorA, temporalVectorB);
}
}
}
void DACE::build_R(double **ax, MyMatrix& R)
{
// takes the array of x vectors, theta, and p, and returns the correlation matrix R.
for(int i=0;i<fCorrelationSize;i++)
{
for(int j=0;j<fCorrelationSize;j++)
{
if(Debug)
fprintf(stdout,"%.5lf %.5lf %.5lf %.5lf %d\n", ax[i][1],ax[j][1],gtheta[1], gp[1], daceSpace->fSearchSpaceDim);
R.insert(i,j,correlation(ax[i+1],ax[j+1], gtheta, gp, fCorrelationSize));
if(Debug)
fprintf(stdout,"%.5lf\n", R(i,j));
}
if(Debug)
fprintf(stdout,"\n");
}
}
bool MyMatrix::operator==(const MyMatrix& a) const
{
if (this->rows() != a.rows()) return false;
if (this->columns() != a.columns()) return false;
return(((EigenMatrix)(*this)-(EigenMatrix)a).isApproxToConstant(0.0));
}
int MyMeshText::CreateString(bool concat, float fontheight, float x, float y, float z, float rotz, unsigned char justificationflags, ColorByte color, Vector2 size, const char* text, ...)
{
assert( m_pFont && m_pFont->m_pFont );
if( strlen( text ) == 0 )
return 0;
const char* stringtodraw = text;
if( g_pLanguageTable != 0 && text[0] == '.' )
stringtodraw = g_pLanguageTable->LookUp( text );
int numlines = 0;
if( concat == false )
{
ClearText();
}
bool moretexttocome = true;
const char* stringpos = stringtodraw;
while( moretexttocome )
{
numlines++;
char singlelinebuffer[300];
singlelinebuffer[0] = 0;
char* singlelinebufferpos = singlelinebuffer;
// word wrap if width of text is not 0.
if( size.x != 0 )
{
float linewidth = -1;// = GetStringSize( fontheight, Vector2(0,0), singlelinebuffer ).x;
while( linewidth < size.x &&
*stringpos != 0 )
{
*singlelinebufferpos = *stringpos;
singlelinebufferpos++;
*singlelinebufferpos = 0;
stringpos++;
linewidth = GetStringSize( fontheight, Vector2(0,0), singlelinebuffer ).x;
assert( singlelinebufferpos < singlelinebuffer + 300 );
}
int numcharswewentback = 0;
while( ( *(singlelinebufferpos-1) != ' ' && *stringpos != 0 ) &&
singlelinebufferpos > singlelinebuffer )
{
singlelinebufferpos--;
numcharswewentback++;
}
if( singlelinebufferpos != singlelinebuffer )
{
*singlelinebufferpos = 0;
stringpos -= numcharswewentback;
}
if( *stringpos == 0 )
moretexttocome = false;
stringtodraw = singlelinebuffer;
}
else
{
moretexttocome = false;
}
//// don't bother drawing if fontheight is zero... still doing logic above so the currect number of lines will be returned.
//if( g_pRTQGlobals->m_WordWrapCountLinesOnly )
// continue;
Vertex_XYZUV_RGBA* pVertsToDraw = (Vertex_XYZUV_RGBA*)GetVerts( true );
int newverts = (int)strlen( stringtodraw ) * 4;
#if _DEBUG
m_MostLettersAttemptedToDrawThisFrame += newverts/4;
if( m_MostLettersAttemptedToDrawThisFrame > m_MostLettersAttemptedToDrawEver )
m_MostLettersAttemptedToDrawEver = m_MostLettersAttemptedToDrawThisFrame;
#endif
if( m_NumVertsToDraw + newverts > GetNumVerts() )
{
#if _DEBUG
LOGInfo( LOGTag, "TextMesh buffer isn't big enough for string (%s) - %d of %d letters used - most letters needed (%d)\n", stringtodraw, m_NumVertsToDraw/4, GetNumVerts()/4, m_MostLettersAttemptedToDrawEver );
#endif
//assert( false ); // drawing more than we have room for.
return 0;
}
pVertsToDraw += m_NumVertsToDraw;
unsigned int textstrlen = m_pFont->m_pFont->GenerateVerts( stringtodraw, true, pVertsToDraw, fontheight, GL_TRIANGLES, justificationflags, color );
m_NumVertsToDraw += (unsigned short)(textstrlen * 4);
m_NumIndicesToDraw += textstrlen * 6;
MyMatrix position;
position.SetIdentity();
position.Rotate( rotz, 0, 0, 1 );
position.SetPosition( x, y - (numlines-1)*fontheight, z );
//position.SetPosition( x, y - (numlines-1)*g_pRTQGlobals->m_WordWrapLineIncSize, z );
//position.SetPosition( x, y, z );
for( unsigned int i=0; i<textstrlen*4; i++ )
{
Vector3 out = position.TransformVector3( *(Vector3*)&pVertsToDraw[i].x );
pVertsToDraw[i].x = out.x;
pVertsToDraw[i].y = out.y;
pVertsToDraw[i].z = out.z;
}
}
return numlines;
}
int main(int argc, char *argv[]){
Params params;
std::map<std::string, std::string> args;
readArgs(argc, argv, args);
if(args.find("algo")!=args.end()){
params.algo = args["algo"];
}else{
params.algo = "qdMCNat";
}
if(args.find("inst_file")!=args.end())
setParamsFromFile(args["inst_file"], args, params);
else
setParams(params.algo, args, params);
createLogDir(params.dir_path);
gen.seed(params.seed);
// Load the dataset
MyMatrix X_train, X_valid;
VectorXd Y_train, Y_valid;
loadMnist(params.ratio_train, X_train, X_valid, Y_train, Y_valid);
//loadCIFAR10(params.ratio_train, X_train, X_valid, Y_train, Y_valid);
//loadLightCIFAR10(params.ratio_train, X_train, X_valid, Y_train, Y_valid);
// ConvNet parameters
std::vector<ConvLayerParams> conv_params;
ConvLayerParams conv_params1;
conv_params1.Hf = 5;
conv_params1.stride = 1;
conv_params1.n_filter = 20;
conv_params1.padding = 0;
conv_params.push_back(conv_params1);
ConvLayerParams conv_params2;
conv_params2.Hf = 5;
conv_params2.stride = 1;
conv_params2.n_filter = 50;
conv_params2.padding = 0;
conv_params.push_back(conv_params2);
std::vector<PoolLayerParams> pool_params;
PoolLayerParams pool_params1;
pool_params1.Hf = 2;
pool_params1.stride = 2;
pool_params.push_back(pool_params1);
PoolLayerParams pool_params2;
pool_params2.Hf = 2;
pool_params2.stride = 2;
pool_params.push_back(pool_params2);
const unsigned n_conv_layer = conv_params.size();
for(unsigned l = 0; l < conv_params.size(); l++){
if(l==0){
conv_params[l].filter_size = conv_params[l].Hf * conv_params[l].Hf * params.img_depth;
conv_params[l].N = (params.img_width - conv_params[l].Hf + 2*conv_params[l].padding)/conv_params[l].stride + 1;
}
else{
conv_params[l].filter_size = conv_params[l].Hf * conv_params[l].Hf * conv_params[l-1].n_filter;
conv_params[l].N = (pool_params[l-1].N - conv_params[l].Hf + 2*conv_params[l].padding)/conv_params[l].stride + 1;
}
pool_params[l].N = (conv_params[l].N - pool_params[l].Hf)/pool_params[l].stride + 1;
}
// Neural Network parameters
const unsigned n_training = X_train.rows();
const unsigned n_valid = X_valid.rows();
const unsigned n_feature = X_train.cols();
const unsigned n_label = Y_train.maxCoeff() + 1;
params.nn_arch.insert(params.nn_arch.begin(),conv_params[n_conv_layer-1].n_filter * pool_params[n_conv_layer-1].N * pool_params[n_conv_layer-1].N);
params.nn_arch.push_back(n_label);
const unsigned n_layers = params.nn_arch.size();
// Optimization parameter
const int n_train_batch = ceil(n_training/(float)params.train_minibatch_size);
const int n_valid_batch = ceil(n_valid/(float)params.valid_minibatch_size);
double prev_loss = std::numeric_limits<double>::max();
double eta = params.eta;
// Create the convolutional layer
std::vector<MyMatrix> conv_W(n_conv_layer);
std::vector<MyMatrix> conv_W_T(n_conv_layer);
std::vector<MyVector> conv_B(n_conv_layer);
// Create the neural network
MyMatrix W_out(params.nn_arch[n_layers-2],n_label);
std::vector<MySpMatrix> W(n_layers-2);
std::vector<MySpMatrix> Wt(n_layers-2);
std::vector<MyVector> B(n_layers-1);
double init_sigma = 0.;
ActivationFunction act_func;
ActivationFunction eval_act_func;
if(params.act_func_name=="sigmoid"){
init_sigma = 4.0;
act_func = std::bind(logistic,true,_1,_2,_3);
eval_act_func = std::bind(logistic,false,_1,_2,_3);
}else if(params.act_func_name=="tanh"){
init_sigma = 1.0;
act_func = std::bind(my_tanh,true,_1,_2,_3);
eval_act_func = std::bind(my_tanh,false,_1,_2,_3);
}else if(params.act_func_name=="relu"){
init_sigma = 1.0; // TODO: Find the good value
act_func = std::bind(relu,true,_1,_2,_3);
eval_act_func = std::bind(relu,false,_1,_2,_3);
}else{
std::cout << "Not implemented yet!" << std::endl;
assert(false);
}
std::cout << "Initializing the network... ";
params.n_params = initNetwork(params.nn_arch, params.act_func_name, params.sparsity, conv_params, pool_params, W_out, W, Wt, B, conv_W, conv_W_T, conv_B); // TODO: Init the conv bias
// Deep copy of parameters for the adaptive rule
std::vector<MyMatrix> mu_dW(n_layers-1);
std::vector<MyVector> mu_dB(n_layers-1);
MyMatrix pW_out = W_out;
std::vector<MySpMatrix> pW = W;
std::vector<MySpMatrix> pWt = Wt;
std::vector<MyVector> pB = B;
MyMatrix ppMii_out, ppM0i_out;
MyVector ppM00_out;
std::vector<MySpMatrix> ppMii,ppM0i;
std::vector<MyVector> ppM00;
MyMatrix pMii_out,pM0i_out;
MyVector pM00_out;
std::vector<MySpMatrix> pMii,pM0i;
std::vector<MyVector> pM00;
std::vector<MyMatrix> conv_ppMii, conv_ppM0i;
std::vector<MyVector> conv_ppM00;
std::vector<MyMatrix> conv_pMii, conv_pM0i;
std::vector<MyVector> conv_pM00;
// Convert the labels to one-hot vector
MyMatrix one_hot = MyMatrix::Zero(n_training, n_label);
labels2oneHot(Y_train,one_hot);
// Configure the logger
std::ostream* logger;
if(args.find("verbose")!=args.end()){
getOutput("",logger);
}else{
getOutput(params.file_path,logger);
}
double cumul_time = 0.;
printDesc(params, logger);
printConvDesc(params, conv_params, pool_params, logger);
std::cout << "Starting the learning phase... " << std::endl;
*logger << "Epoch Time(s) train_loss train_accuracy valid_loss valid_accuracy eta" << std::endl;
for(unsigned i = 0; i < params.n_epoch; i++){
for(unsigned j = 0; j < n_train_batch; j++){
// Mini-batch creation
unsigned curr_batch_size = 0;
MyMatrix X_batch, one_hot_batch;
getMiniBatch(j, params.train_minibatch_size, X_train, one_hot, params, conv_params[0], curr_batch_size, X_batch, one_hot_batch);
double prev_time = gettime();
// Forward propagation for conv layer
std::vector<std::vector<unsigned>> poolIdxX1(n_conv_layer);
std::vector<std::vector<unsigned>> poolIdxY1(n_conv_layer);
MyMatrix z0;
std::vector<MyMatrix> conv_A(conv_W.size());
std::vector<MyMatrix> conv_Ap(conv_W.size());
convFprop(curr_batch_size, conv_params, pool_params, act_func, conv_W, conv_B, X_batch, conv_A, conv_Ap, z0, poolIdxX1, poolIdxY1);
// Forward propagation
std::vector<MyMatrix> Z(n_layers-1);
std::vector<MyMatrix> A(n_layers-2);
std::vector<MyMatrix> Ap(n_layers-2);
fprop(params.dropout_flag, act_func, W, W_out, B, z0, Z, A, Ap);
// Compute the output and the error
MyMatrix out;
softmax(Z[n_layers-2], out);
std::vector<MyMatrix> gradB(n_layers-1);
gradB[n_layers-2] = out - one_hot_batch;
// Backpropagation
bprop(Wt, W_out, Ap, gradB);
// Backpropagation for conv layer
std::vector<MyMatrix> conv_gradB(conv_W.size());
MyMatrix layer_gradB = (gradB[0] * W[0].transpose());
MyMatrix pool_gradB;
layer2pool(curr_batch_size, pool_params[conv_W.size()-1].N, conv_params[conv_W.size()-1].n_filter, layer_gradB, pool_gradB);
convBprop(curr_batch_size, conv_params, pool_params, conv_W_T, conv_Ap, pool_gradB, conv_gradB, poolIdxX1, poolIdxY1);
if(params.algo == "bprop"){
update(eta, gradB, A, z0, params.regularizer, params.lambda, W_out, W, Wt, B);
convUpdate(curr_batch_size, eta, conv_params, conv_gradB, conv_A, X_batch, "", 0., conv_W, conv_W_T, conv_B);
}else{
// Compute the metric
std::vector<MyMatrix> metric_gradB(n_layers-1);
std::vector<MyMatrix> metric_conv_gradB(conv_params.size());
if(params.algo=="qdMCNat"){
// Monte-Carlo Approximation of the metric
std::vector<MyMatrix> mc_gradB(n_layers-1);
computeMcError(out, mc_gradB[n_layers-2]);
// Backpropagation
bprop(Wt, W_out, Ap, mc_gradB);
for(unsigned k = 0; k < gradB.size(); k++){
metric_gradB[k] = mc_gradB[k].array().square();
}
// Backpropagation for conv layer
std::vector<MyMatrix> mc_conv_gradB(conv_W.size());
MyMatrix mc_layer_gradB = (mc_gradB[0] * W[0].transpose());
MyMatrix mc_pool_gradB;
layer2pool(curr_batch_size, pool_params[conv_W.size()-1].N, conv_params[conv_W.size()-1].n_filter, mc_layer_gradB, mc_pool_gradB);
convBprop(curr_batch_size, conv_params, pool_params, conv_W_T, conv_Ap, mc_pool_gradB, mc_conv_gradB, poolIdxX1, poolIdxY1);
for(unsigned k = 0; k < conv_params.size(); k++){
metric_conv_gradB[k] = mc_conv_gradB[k].array().square();
}
}
else if(params.algo=="qdop"){
for(unsigned k = 0; k < conv_params.size(); k++){
metric_conv_gradB[k] = conv_gradB[k].array().square();
}
for(unsigned k = 0; k < gradB.size(); k++){
metric_gradB[k] = gradB[k].array().square();
}
}
else if(params.algo=="qdNat"){
for(unsigned k = 0; k < conv_params.size(); k++){
metric_conv_gradB[k] = conv_gradB[k].array().square();
}
for(unsigned k = 0; k < metric_gradB.size(); k++){
metric_gradB[k] = MyMatrix::Zero(gradB[k].rows(),gradB[k].cols());
}
for(unsigned l = 0; l < n_label; l++){
MyMatrix fisher_ohbatch = MyMatrix::Zero(curr_batch_size, n_label);
fisher_ohbatch.col(l).setOnes();
std::vector<MyMatrix> fgradB(n_layers-1);
fgradB[n_layers-2] = out - fisher_ohbatch;
bprop(Wt, W_out, Ap, fgradB);
// Backpropagation for conv layer
std::vector<MyMatrix> fisher_conv_gradB(conv_W.size());
MyMatrix fisher_layer_gradB = (fgradB[0] * W[0].transpose());
MyMatrix fisher_pool_gradB;
layer2pool(curr_batch_size, pool_params[conv_W.size()-1].N, conv_params[conv_W.size()-1].n_filter, fisher_layer_gradB, fisher_pool_gradB);
convBprop(curr_batch_size, conv_params, pool_params, conv_W_T, conv_Ap, fisher_pool_gradB, fisher_conv_gradB, poolIdxX1, poolIdxY1);
for(unsigned k = 0; k < conv_params.size(); k++){
MyMatrix fisher_conv_gradB_sq = fisher_conv_gradB[k].array().square();
for(unsigned m = 0; m < out.rows(); m++){
for(unsigned f = 0; f < conv_params[k].n_filter; f++){
for(unsigned n = 0; n < conv_params[k].N * conv_params[k].N; n++){
fisher_conv_gradB_sq(f,m*conv_params[k].N*conv_params[k].N+n) *= out(m,l);
}
}
}
metric_conv_gradB[k] += fisher_conv_gradB_sq;
}
for(unsigned k = 0; k < W.size(); k++){
const unsigned rev_k = n_layers - k - 2;
metric_gradB[rev_k] += (fgradB[rev_k].array().square().array().colwise() * out.array().col(l)).matrix();
}
}
}
bool init_flag = false;
if(i == 0 && j == 0 && !params.init_metric_id){
init_flag = true;
}
std::vector<MyMatrix> conv_Mii(conv_params.size());
std::vector<MyMatrix> conv_M0i(conv_params.size());
std::vector<MyVector> conv_M00(conv_params.size());
buildConvQDMetric(curr_batch_size, metric_conv_gradB, conv_A, X_batch, conv_W, params.matrix_reg, conv_Mii, conv_M0i, conv_M00);
updateConvMetric(init_flag, params.metric_gamma, conv_pMii, conv_pM0i, conv_pM00, conv_Mii, conv_M0i, conv_M00);
MyMatrix Mii_out, M0i_out;
MyVector M00_out;
std::vector<MySpMatrix> Mii(W.size());
std::vector<MySpMatrix> M0i(W.size());
std::vector<MyVector> M00(W.size());
buildQDMetric(metric_gradB, A, z0, W_out, W, params.matrix_reg, Mii_out, M0i_out, M00_out, Mii, M0i, M00);
updateMetric(init_flag, params.metric_gamma, Mii_out, M0i_out, M00_out, Mii, M0i, M00, pMii_out, pM0i_out, pM00_out, pMii, pM0i, pM00);
update(eta, gradB, A, z0, params.regularizer, params.lambda, W_out, W, Wt, B, Mii_out, M0i_out, M00_out, Mii, M0i, M00);
}
double curr_time = gettime();
cumul_time += curr_time - prev_time;
if(params.minilog_flag){
double train_loss = 0.;
double train_accuracy = 0.;
double valid_loss = 0.;
double valid_accuracy = 0.;
evalModel(eval_act_func, params, n_train_batch, n_training, X_train, Y_train, conv_params, pool_params, conv_W, conv_B, W_out, W, B, train_loss, train_accuracy);
evalModel(eval_act_func, params, n_valid_batch, n_valid, X_valid, Y_valid, conv_params, pool_params, conv_W, conv_B, W_out, W, B, valid_loss, valid_accuracy);
// Logging
*logger << i + float(j)/n_train_batch << " " << cumul_time << " " << train_loss << " " << train_accuracy << " " << valid_loss << " " << valid_accuracy << " " << eta << std::endl;
}
}
if(!params.minilog_flag || params.adaptive_flag){
double train_loss = 0.;
double train_accuracy = 0.;
double valid_loss = 0.;
double valid_accuracy = 0.;
evalModel(eval_act_func, params, n_train_batch, n_training, X_train, Y_train, conv_params, pool_params, conv_W, conv_B, W_out, W, B, train_loss, train_accuracy);
evalModel(eval_act_func, params, n_valid_batch, n_valid, X_valid, Y_valid, conv_params, pool_params, conv_W, conv_B, W_out, W, B, valid_loss, valid_accuracy);
// if(params.adaptive_flag)
// adaptiveRule(train_loss, prev_loss, eta, W, B, pMii, pM0i, pM00, pW, pB, ppMii, ppM0i, ppM00);
// Logging
if(!params.minilog_flag){
*logger << i << " " << cumul_time << " " << train_loss << " " << train_accuracy << " " << valid_loss << " " << valid_accuracy << " " << eta << std::endl;
}
}
}
}
void createHamiltonian(Geometry const &geometry, DynVars const &dynVars,
MyMatrix<std::complex<double> >& matrix,Parameters const ðer,Aux &aux,int type)
// Modified by IS Nov-08-04
{
int col, volume, i = 0, p,dir;
tpem_t hopping,hopping2,bandHop;
double tmp,tmp2;
int j,iTmp;
volume = ether.linSize;
tpem_t S_ij;
//static vector<double> phonon_q1(volume);
//static vector<double> phonon_q2(volume);
//static vector<double> phonon_q3(volume);
Phonons<Parameters,Geometry> phonons(ether,geometry);
for (p = 0; p < matrix.getRank(); p++)
for (col = 0; col < matrix.getRank(); col++)
matrix(p,col)=0;
for (p = 0; p < volume; p++) {
double phonon_q1=phonons.calcPhonon(p,dynVars,0);
double phonon_q2=phonons.calcPhonon(p,dynVars,1);
double phonon_q3=phonons.calcPhonon(p,dynVars,2);
matrix(p,p) = ether.phononEjt[0]*phonon_q1+ether.phononEjt[2]*phonon_q3+ether.potential[p];
matrix(p+volume,p+volume) = -ether.phononEjt[2]*phonon_q3+ether.phononEjt[0]*phonon_q1+ether.potential[p];
matrix(p,p+volume) = (ether.phononEjt[1]*phonon_q2);
matrix(p+volume,p) = conj(matrix(p,p+volume));
for (j = 0; j < geometry.z(p); j++) { /* hopping elements */
iTmp=geometry.borderId(p,j);;
if (iTmp>=0) hopping = ether.hoppings[iTmp];
else hopping= -1.0;
col = geometry.neighbor(p,j);
tmp=cos(0.5*dynVars.theta[p])*cos(0.5*dynVars.theta[col]);
tmp2=sin(0.5*dynVars.theta[p])*sin(0.5*dynVars.theta[col]);
S_ij=tpem_t(tmp+tmp2*cos(dynVars.phi[p]-dynVars.phi[col]),
-tmp2*sin(dynVars.phi[p]-dynVars.phi[col]));
if (p>col) hopping2 = conj(hopping);
else hopping2=hopping;
dir = int(j/2);
bandHop=ether.bandHoppings[0+0*2+dir*4];
hopping=hopping2 * bandHop ;
matrix(p,col) = hopping * S_ij;
matrix(col,p) = conj(hopping * S_ij);
bandHop=ether.bandHoppings[0+1*2+dir*4];
hopping= hopping2 * bandHop;
matrix(p, col+volume)=hopping * S_ij;
matrix(col+volume,p)=conj(matrix(p,col+volume));
//
bandHop=ether.bandHoppings[1+0*2+dir*4];
hopping= hopping2 * bandHop;
matrix(p+volume,col) = hopping * S_ij;
matrix(col,p+volume) = conj(matrix(p+volume,col));
bandHop=ether.bandHoppings[1+1*2+dir*4];
hopping=hopping2 * bandHop;
matrix(p+volume,col+volume) = hopping * S_ij;
matrix(col+volume,p+volume) = conj(matrix(p+volume,col+volume));
}
if (ether.isSet("tprime")) {
for (j = 0; j < geometry.z(p,2); j++) { /* hopping elements */
iTmp=geometry.borderId(p,j,2);
hopping= -1.0;
if (iTmp>=0) hopping = ether.hoppings[iTmp];
col = geometry.neighbor(p,j,2);
tmp=cos(0.5*dynVars.theta[p])*cos(0.5*dynVars.theta[col]);
tmp2=sin(0.5*dynVars.theta[p])*sin(0.5*dynVars.theta[col]);
S_ij=tpem_t(tmp+tmp2*cos(dynVars.phi[p]-dynVars.phi[col]),
-tmp2*sin(dynVars.phi[p]-dynVars.phi[col]));
if (p>col) hopping2 = conj(hopping);
else hopping2=hopping;
dir = int(j/2);
bandHop=ether.bandHoppings[0+0*2+dir*4]*ether.tprime;
hopping=hopping2 * bandHop ;
matrix(p,col) = hopping * S_ij;
matrix(col,p) = conj(hopping * S_ij);
bandHop=ether.bandHoppings[0+1*2+dir*4]*ether.tprime;
hopping= hopping2 * bandHop;
matrix(p, col+volume)=hopping * S_ij;
matrix(col+volume,p)=conj(matrix(p,col+volume));
//
bandHop=ether.bandHoppings[1+0*2+dir*4]*ether.tprime;
hopping= hopping2 * bandHop;
matrix(p+volume,col) = hopping * S_ij;
matrix(col,p+volume) = conj(matrix(p+volume,col));
bandHop=ether.bandHoppings[1+1*2+dir*4]*ether.tprime;
hopping=hopping2 * bandHop;
matrix(p+volume,col+volume) = hopping * S_ij;
matrix(col+volume,p+volume) = conj(matrix(p+volume,col+volume));
}
}
}
//if (!matrix.isHermitian()) throw std::runtime_error("I'm barking\n");
}