PLearn 0.1
Namespaces | Functions
plapack.cc File Reference
#include <cstdlib>
#include "plapack.h"
#include <algorithm>
#include "random.h"
Include dependency graph for plapack.cc:

Go to the source code of this file.

Namespaces

namespace  PLearn
 

< for swap


Functions

int PLearn::eigen_SymmMat (Mat &in, Vec &e_value, Mat &e_vector, int &n_evalues_found, bool compute_all, int nb_eigen, bool compute_vectors, bool largest_evalues)
int PLearn::eigen_SymmMat_decreasing (Mat &in, Vec &e_value, Mat &e_vector, int &n_evalues_found, bool compute_all, int nb_eigen, bool compute_vectors=true, bool largest_evalues=true)
 same as the previous call, but eigenvalues/vectors are sorted by largest firat (in decreasing order)
int PLearn::matInvert (Mat &in, Mat &inverse)
 This function compute the inverse of a matrix.
int PLearn::lapackSolveLinearSystem (Mat &At, Mat &Bt, TVec< int > &pivots)
void PLearn::solveLinearSystem (const Mat &A, const Mat &Y, Mat &X)
 for matrices A such that A.length() <= A.width(), find X s.t.
void PLearn::solveTransposeLinearSystem (const Mat &A, const Mat &Y, Mat &X)
 for matrices A such that A.length() >= A.width(), find X s.t.
Mat PLearn::solveLinearSystem (const Mat &A, const Mat &B)
Vec PLearn::solveLinearSystem (const Mat &A, const Vec &b)
 Returns solution x of Ax = b (same as above, except b and x are vectors)
Vec PLearn::constrainedLinearRegression (const Mat &Xt, const Vec &Y, real lambda)
void PLearn::lapackCholeskyDecompositionInPlace (Mat &A, char uplo='L')
 Call LAPACK to perform in-place Cholesky Decomposition of a square SYMMETRIC matrix A.
void PLearn::lapackCholeskySolveInPlace (Mat &A, Mat &B, bool B_is_column_major=false, char uplo='L')
 Call LAPACK to solve in-place a linear system given its previously-computed Cholesky decomposition.
Mat PLearn::multivariate_normal (const Vec &mu, const Mat &A, int N)
 generate N vectors sampled from the normal with mean vector mu and covariance matrix A
Vec PLearn::multivariate_normal (const Vec &mu, const Mat &A)
 generate a vector sampled from the normal with mean vector mu and covariance matrix A
Vec PLearn::multivariate_normal (const Vec &mu, const Vec &e_values, const Mat &e_vectors)
 generate 1 vector sampled from the normal with mean mu and covariance matrix A = evectors * diagonal(e_values) * evectors'
void PLearn::multivariate_normal (Vec &x, const Vec &mu, const Vec &e_values, const Mat &e_vectors, Vec &z)
 generate a vector x sampled from the normal with mean mu and covariance matrix A = evectors * diagonal(e_values) * evectors' (the normal(0,I) originally sampled to obtain x is stored in z).
void PLearn::affineNormalization (Mat data, Mat W, Vec bias, real regularizer)
real PLearn::GCV (Mat X, Mat Y, real weight_decay, bool X_is_transposed, Mat *W)
 Compute the generalization error estimator called Generalized Cross-Validation (Craven & Wahba 1979), and the corresponding ridge regression weights in min ||Y - X*W'||^2 + weight_decay ||W||^2.
real PLearn::GCVfromSVD (real n, real Y2minusZ2, Vec Z, Vec s)
 Estimator of generalization error estimator called Generalized Cross-Validation (Craven & Wahba 1979), computed from the SVD of the input matrix X in the ridge regression.
real PLearn::ridgeRegressionByGCV (Mat X, Mat Y, Mat W, real &best_GCV, bool X_is_transposed=false, real initial_weight_decay_guess=-1, int explore_threshold=5, real min_weight_decay=0)
 Perform ridge regression WITH model selection (i.e.
real PLearn::weightedRidgeRegressionByGCV (Mat X, Mat Y, Vec gamma, Mat W, real &best_gcv, real min_weight_decay=0)
 Similar to ridgeRegressionByGCV, but with support form sample weights gamma.
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Defines