\begin{verbatim}
README file for cocluster
USAGE
cocluster-'uname' [SWITCHES]
DESCRIPTION
Implements three co-clustering algorithms (in C++):
information-theoretic co-clustering and two types of
minimum sum-squared residue co-clustering algorithms
AUTHORS
Hyuk Cho
Copyright (c) 2005, 2006
Hyuk Cho, Yuqiang Guan, and Suvrit Sra
{hyukcho, yguan, suvrit}@cs.utexas.edu
Copyright (c) 2003, 2004
REQUIRED SWITCHES
At least, the following switches are required: -A -C -I -R
DESCRIPTION OF SWITCHES
-A algorithmType (REQUIRED)
specifies type of coclustering algorithm
e -- euclidean coclustering algorithm (DEFAULT)
i -- information theoretic coclustering algorithm
r -- minimim squared residue coclustering algorithm
-C colClusterNum (REQUIRED)
specifies number of column clusters (DEFAULT = 1)
-D dumpLevel [dumpAccessMode dumpFilename]
specifies level of intermediate information dump
dumpLevel can be one of
0 -- minimal information (DEFAULT)
1 -- objective function value in Batch Update and minimal statistics
2 -- objective function value in local search and maximal statistics
3 -- maximal information in a specified file and minimal on stdout
dumpAccessMode is required after '-D 3' and one of
a -- append mode
o -- output mode
dumpFilename is required after '-D 3 a/o'
-E outputEachClsuter
specifies detail ouputs in external cluster validattion section
outputEachCluster can be one of
0 -- don't output a confusion matrix and each cluster's statistics (DEFAULT)
1 -- output them in external cluster validation section
-I inputMatrixType inputFormatType inputFilename (REQUIRED)
specifies details of input matrix file
inputMatrixType can be one of
d -- dense (i.e., rectangle/square) matrix
inputFormatType after '-I d' can be one of
s -- dimension and matrix stored separately
t -- both dimension and matrix stored together
s -- sparse matrix in CCS
inputFormatType after '-I s' can be one of
f -- tfn scaling
x -- txx scaling
-J oneWay
specifies to get one-way row/column clustering objective function values
oneWay can be one of
0 -- don't compute one-way objective function value(s) (DEFAULT)
1 -- compute one-way objective function value(s)
-K classLabelSelection classOffsetType classLabelFilename
specifies details of class label file
classLabelSelection can be one of
b -- both row and column class labels
c -- column class labels
r -- row class labels
classOffsetType can be one of
0 -- class label index starts from 0
1 -- class label index starts from 1
-M smoothingType [rowSmoothingMagnitude colSmoothingMagnitude]
specifies details of smoothing techniques
smoothingType can be one of
a -- annealing of uniform smoothing (DEFAULT = 1)
h -- maximum entropy smoothing
n -- no smoothing (DEFAULT)
u -- uniform smoothing (DEFAULT = 0)
smoothingMagnitude is required for '-M a/h/u'
-N runNum
specifies number of runs to get averaged statistics (DEFAULT = 1)
-O outputFileType [outputLabelType outputOffsetType] outputAccessMode outputFilename
specifies details of cocluster/objective/statistics file information
outputFileType can be one of
c -- output co-clusters
outputLabelType is required after '-O c' and can be one of
b -- each co-cluster block consisting of three rows like
#rows #columns (in 1st row)
list of row #s in (1,1)-th co-cluster (in 2nd row)
list of column #s in (1,2)-th column co-cluster (in 3rd row)
and so on
s -- all co-clusters represented in a simple format of two rows like (DEFAULT)
list of row cluster labels of all rows (in 1st row)
list of column cluster labels of all columns (in 2nd row)
outputOffsetType is required after '-O c b' and outputLabelType and can be one of
0 -- cluster label index starts from 0
1 -- cluster label index starts from 1
o -- output objective function value(s)
s -- output satistical information
outputAccessMode can be one of
a -- append mode, useful with '-N randomRunNum'
o -- output mode
-R rowClusterNum (REQUIRED)
specifies number of row clusters (DEFAULT = 1)
-S seedingType seedingLabelSelection [seedingPertValue
| (seedingOffsetType numSeedingSet seedingFilename)
| (numRowPermutation | numColPermutation)]
specifies details of initial cluster assignment
seedingType can be one of
f -- farthest apart assignment
m -- permute an initial random cluster vector
numRowPermutation or/and numColPermutation is/are required after '-S m b/c/r'
numRowPermutation numColPermutation -- after '-S m b'
numColPermutation -- after '-S m c'
numRowPermutation -- after '-S m r'
p -- perturbates cluster centroids
seedingPertValue is required after '-S p'
r -- random assignment (DEFAULT)
s -- read cluster labels from a seeding file
seedingOffsetType is required after '-S s b/c/r' and one of
0 -- seeding label index starts from 0
1 -- seeding label index starts from 1
numSeedingSet is required after '-S s b/c/r 0/1'
n -- positive integer (>=1)
seedingFilename is required after '-S s b/c/r/ 0/1 n'
seedingLabelSelection can be one of
b -- both row and column cluster labels
c -- column cluster labels
r -- row cluster labels
-T thresholdType rowThreshold colThreshold
specifies details of threshold for either batch update or local search
thresholdType can be one of
b -- batch update (default = +0.001)
l -- local search (default = -1e-06)
Both rowThreshold and colThreshold are required.
-U updateType [updateSelection | (rowLocalSearchLength colLocalSearchLength)]
specifies details of cluster-centroid-updating order in batch update or local search
updateType can be one of
b -- batch update
updateSelection is required after '-U b' and can be one of
0 -- single row and single column, respectively (DEFAULT)
1 -- single row and single column, in batch mode
2 -- single row or single column, flipping a pair coin
3 -- multiple run of either row or column, flipping a pair coin
l -- local search
rowLocalSearchLength (DEFAULT = 0) is required after 'l'
colLocalSearchLength (DEFAULT = 0) is required after 'l' and rowLocalSearchLength
To avoid empty row/column cluster(s), use -1 for row/colLocalSearchLength.
-X anticorrelation
specifies to capture anti-correlated rows (by taking reverse of rows)
anticorrelation can be one of
0 -- don't capture anti-correlated rows (DEFAULT)
1 -- capture anti-correlated rows
\end{verbatim}