comparison toolboxes/alps/generate_vector.m @ 154:0de08f68256b ivand_dev

ALPS toolbox - Algebraic Pursuit added to smallbox

+function [x] = generate_vector(N, K, ensemble, sigma, p)
+% =========================================================================
+%                       Sparse vector generator
+% =========================================================================
+% INPUT ARGUMENTS:
+% N                         Size of sparse vector.
+% K                         Sparsity of vector.
+% ensemble                  Ensemble type of measurement matrix. Possible
+%                           values are:
+%                           -'Gaussian': K non-zero elements of the sparse vector
+%                           are drawn from normal distribution N(0,1).
+%                           -'sGaussian': K non-zero elements of the sparse vector
+%                           are drawn from normal distribution N(0,sigma^2).
+%                           -'Bernoulli': K non-zero elements of the sparse vector
+%                           are drawn from Bernoulli distribution (1/2,1/2).
+%                           -'pBernoulli': K non-zero elements of the sparse vector
+%                           are drawn from Bernoulli distribution (p,1-p).
+% sigma                     Standard deviation of Gaussian distribution.
+% p                         Parameter of Bernoulli distribution.
+% =========================================================================
+% OUTPUT ARGUMENTS:
+% x                     K-sparse vector.
+% =========================================================================
+% 01/04/2011, by Anastasios Kyrillidis. [e-mail], EPFL.
+% =========================================================================
+
+if nargin < 3
+    ensemble = 'Gaussian';
+end;
+
+if nargin < 4
+    sigma = 1;
+    p = 0.5; 
+end;
+
+x = zeros(N,1);
+rand_indices = randperm(N);
+
+switch ensemble
+    case 'Gaussian'
+        x(rand_indices(1:K)) = randn(K,1);         % Standard normal distribution ~ N(0,1)
+    case 'sGaussian'
+        x(rand_indices(1:K)) = sigma*randn(K,1);    % Normal distribution ~ N(0,sigma^2)
+    case 'Uniform'
+        x(rand_indices(1:K)) = rand(K,1);           % Uniform [0,1] distribution
+    case 'Bernoulli'
+        x(rand_indices(1:K)) = (-1).^round(rand(K,1));     % Bernoulli ~ (1/2, 1/2) distribution
+    case 'pBernoulli'
+        x(rand_indices(1:K)) = (-1).^(rand(K,1) > p);   % Bernoulli ~ (p, 1-p) distribution     
+end;
+
+x = x/norm(x);