void histoBook::make( xmlConfig * config, string nodeName ){
if ( config && config->nodeExists( nodeName ) ){
string hName = config->tagName( nodeName );
if ( "" == hName )
hName = nodeName;
// store the path in the config file
configPath[ hName ] = nodeName;
string type = config->getString( nodeName + ":type", "1D" );
if ( "1D" == type ){
make1D( hName, config->getString( nodeName + ":title", hName ),
config->getInt( nodeName + ":nBinsX", 1 ), config->getDouble( nodeName + ":x1", 0 ),
config->getDouble( nodeName + ":x2", 1 ) );
} else if ( "2D" == type ){
make2D( hName, config->getString( nodeName + ":title", hName ),
config->getInt( nodeName + ":nBinsX", 1 ), config->getDouble( nodeName + ":x1", 0 ),
config->getDouble( nodeName + ":x2", 1 ),
config->getInt( nodeName + ":nBinsY", 1 ), config->getDouble( nodeName + ":y1", 0 ),
config->getDouble( nodeName + ":y2", 1 ) );
}
}
}
// Applies previously trained QDA classifier to new data matrix x
//
// Inputs:
// i_x - N x p data matrix of N samples in p dimensions
// i_means - K x p matrix of class centroids
// i_icovmats - K x p x p array of inverse covariance matrices
// i_bias - vector of length containing the bias term for each discriminant function
// i_N - number of samples in x
// i_p - dimensionality of x
// i_K - number of classes
//
// Outputs:
// o_labels - labels for all samples in i_x
//
// Memory allocation requirements for the outputs:
// o_labels must be of length N
void QDAtest(
double* i_x, double* i_means, double* i_icovmats, double* i_bias,
int* i_N, int* i_p, int* i_K, int* o_labels
)
{
// Create local storage
int t, k, i, j;
double temp;
double* x_c = make1D( *i_p );
double** x = make2D( *i_N, *i_p );
double** means = make2D( *i_K, *i_p );
double*** icovmats = make3D( *i_K, *i_p, *i_p );
double** delta = make2D( *i_N, *i_K );
// Copy and reshape the inputs to local storage
cp_vec_mat( *i_N, *i_p, i_x, x );
cp_vec_mat( *i_K, *i_p, i_means, means );
cp_vec_3D( *i_K, *i_p, *i_p, i_icovmats, icovmats );
// Compute the delta function values
for( t = 0; t < *i_N; t++ )
for( k = 0; k < *i_K; k++ )
{
add_vec_vec( *i_p, 1.0, -1.0, x[t], means[k], x_c );
delta[t][k] = 0.0;
for( i = 0; i < *i_p; i++ )
for( j = 0; j < *i_p; j++ )
delta[t][k] += icovmats[k][i][j] * x_c[i] * x_c[j];
delta[t][k] *= -0.5;
delta[t][k] += i_bias[k];
}
// Pick the highest discriminant function value for each point
for( t = 0; t < *i_N; t++ )
{
o_labels[t] = 1;
temp = delta[t][0];
for( k = 1; k < *i_K; k++ )
{
if( delta[t][k] > temp )
{
o_labels[t] = k + 1;
temp = delta[t][k];
}
}
}
// Free up memory
free( x_c );
free2D( x, *i_N );
free2D( means, *i_K );
free3D( icovmats, *i_K, *i_p );
free2D( delta, *i_N );
}
