wanirepo

function L = SBM_likelihood(A, z, varargin)

% usage: L = SBM_likelihood(A, z)
%
% feature: log-likelihood estimation for simple Stochastic Block Model (SBM)
%          (undirected, unweighted) based on Maximum Likelihood estimation. 
%
% input:   A      adjacency matrix (simple graph)
%          z      group membership
%

wanirepo

function [bestL, bestP, info] = KL_heuristic_k2(A, varargin)

% usage: [bestL, bestP, info] = KL_heuristic_k2(A, varargin)
%
% feature: use the Kernighan-Lin (KL) heuristic to optimize any partition 
%          score function, e.g., modularity Q or stochastic block model's
%          likelihood function. This works only for 2 partitioning problem.
%
% input:   A        adjacency matrix
%

wanirepo

cd('/Users/clinpsywoo/github/fastRPCA');
setup_fastRPCA;

% load an example data
load('/Users/clinpsywoo/github/SAS2015_PatRec/data.mat');

X = dat.dat;
SS = svd(X, 'econ');  % can decompose x or x', but faster when rows >> cols
                      % 3 solvers (5 variants)
                      %   1) constrained

wanirepo

poscm = colormap_tor([0.96 0.41 0], [1 1 0]);  % warm
negcm = colormap_tor([0.11 0.46 1], [.23 1 1]);  % cools

cluster_surf(cl ,which('surf_BrainMesh_ICBM152Right_smoothed.mat'), 2, 'heatmap', 'colormaps', poscm, negcm)

axis vis3d;

wanirepo

% In this simulation we examined how reliable peak distance and pattern correlation
% by comparing two simulated data with the same ground truth patterns of signal.

% define SNR levels

snr_all = [.1 .3 .5 .7 .9 1.1];

% create ground truth pattern

pattern = [0.4 1.7 0.5 0.7 0.2