-

Author: Burak Bayramli

algs/mstseg/felz/.gitignore

@@ -0,0 +1 @@
+*.pyc

algs/mstseg/felz/COPYING

@@ -0,0 +1,70 @@
+### Felzenszwalb Clustering
+
+Using a distance matrix containing pairwise distances (which could use
+any measure) Felzenszwalb Clustering will print cluster membership per
+datapoint. 
+
+The code is based on Felzenszwalb and Huttenlocher paper and shared
+C++ code. I simply took out the clustering code from inside the
+segmentation, so this project does not have any references to image
+processing, it simply does clustering with any given distance matrix.
+
+http://cs.brown.edu/~pff/segment/
+
+The C++ version reads sparse matrices from mtx (money market
+format). Python version can be used as any clustering algorithm found
+in `scikit-learn`, through constructor and `fit` function.
+
+The main advantage of the algorithm is its ability to use any distance
+metric, its accuracy (as known from Felzenszwalb image segmentation),
+its speed, and determining the number of clusters on its own. The
+algorithm which is at the root of Felzenszwalb clustering is Minimum
+Spanning Tree which is known to have O(E log E) complexity where E is
+the number of edges in a graph. In comparison, other clustering
+methods presented here,
+
+http://scikit-learn.org/stable/modules/clustering.html
+
+such as Affinity Propagation, or Spectral Clustering do not enjoy the
+same level of performance and cannot handle large datasets.
+
+
+## Building
+
+`cd felzclust`
+
+`make`
+
+For Python version simple import is sufficient.
+
+## Running
+
+See `doc/test.pdf` or `felzclust/test.py` for details. The distance
+matrix should be in sparse form, in mtx format. A simple use case
+
+`felzclust/felzclust felzclust/simple.mtx 1.0 1`
+
+You will see output
+
+```
+point;cluster
+0;2
+1;2
+2;2
+3;4
+4;4
+```
+
+A bigger example
+
+![](./doc/test_02.png)
+
+![](./doc/test_01.png)
+
+## Pure Python
+
+Pure Python version of the code can be found under `felz.py` and `test_felz.py`
+
+## LICENSE
+
+The software is licensed under GPL v2. See COPYING for details.

algs/mstseg/felz/README.md

@@ -0,0 +1,4 @@
+*.aux
+*.log
+*.out
+_region*

algs/mstseg/felz/doc/.gitignore

@@ -0,0 +1,35 @@
+\usepackage{palatino,eulervm}
+\usepackage{showkeys}\renewcommand*\showkeyslabelformat[1]{(#1)}
+\usepackage{anysize}
+\marginsize{3cm}{3cm}{1cm}{3cm}
+\setlength{\parindent}{0pt}
+\usepackage{graphicx}
+\usepackage{cancel}
+\usepackage[latin5]{inputenc}
+\usepackage{color}
+\setlength{\parskip}{8pt}
+\setlength{\parsep}{0pt}
+\setlength{\headsep}{0pt}
+\setlength{\topskip}{0pt}
+\setlength{\topmargin}{0pt}
+\setlength{\topsep}{0pt}
+\setlength{\partopsep}{0pt}
+\setlength{\mathindent}{0cm}
+\usepackage{latexsym}
+\usepackage{amsfonts}
+\usepackage{mathrsfs}
+\definecolor{lbcolor}{rgb}{0.9,0.9,0.9}  
+\usepackage{minted}
+\usepackage{marginnote}
+\newcommand*\mlabel[1]{\marginnote{(#1)}}  
+\usepackage[hyphens]{url}
+\usepackage{hyperref}
+\makeatletter \renewcommand\verbatim@font{\footnotesize\ttfamily} \makeatother
+% pseudocode
+\usepackage{algorithm}
+\newcommand{\code}[1]{\mbox{\texttt{#1}}}
+\newcounter{lineno}
+\newenvironment{pseudocode}{\begin{small}\begin{tabbing}\textbf{mm}\=mm\=mm\=mm\=mm\=mm\=mm\=mm\=mm\=\kill}{\end{tabbing}\end{small}}
+\newcommand{\codename}{\setcounter{lineno}{0}\>}
+\newcommand{\codeline}{\>\stepcounter{lineno}\textbf{\arabic{lineno}}\'\>}
+\newcommand{\codeskip}{\>\>}

algs/mstseg/felz/doc/common.sty

@@ -0,0 +1,2 @@
+pdflatex -shell-escape test.tex
+evince test.pdf

algs/mstseg/felz/doc/doc.sh

@@ -0,0 +1,18 @@
+Simple data generation
+
+\begin{minted}[fontsize=\footnotesize]{python}
+import scipy.sparse as sps
+from scipy.io import mmwrite, mmread
+A = sps.lil_matrix((5,5))
+A[0,1] = 0.5
+A[1,0] = 0.5
+A[1,2] = 1.0
+A[2,1] = 1.0
+A[2,3] = 0.33
+A[3,2] = 0.33
+A[0,4] = 0.5
+A[4,0] = 0.5
+mmwrite('./felzclust/simple.mtx',A)
+\end{minted}
+
+