readme.txt

数据挖掘经典的hierarchial clustering algorithm

How to use BIRCH?

0. Setup of directories:

please set variables in "make_include" file correctly correctly

BIRCH: top directory, "make all" in this directory will create
       "birch" which is the executable. Under this directory
       there are two sub-directories:

   AttrProj: code relevant to integration with DEVISE
          "make all" in this directory will create AttrProjLib.o
   Birch: code relevant to BIRCH;
          "make all" in this directory will create BirchLib.o

1. Command format: birch parafile schemefile projfile datafile

2. Input files:

parafile: BIRCH parameter setting.
          Its meaning is explained in file "para.config" and
          "rg2d.para" is a sample parafile for running BIRCH
          on the data file "rg2d.dat".
          Please look "para.config": the parameters that you might
          need to set according to your data set are marked by "*",
          and all other parameters you can just set as given in it.
          Also please look "rg2d.para" to get a feel of it.
schemefile: Scheme description of data.
          "rg2d.scheme" is a sample schemefile for running
          BIRCH on the data file "rg2d.dat".
projfile: Projection description of data for clustering.
          "rg2d.proj" is a sample projfile for running BIRCH
          on the data file "rg2d.dat" and do clustering on
          attribute X, Y separately, and X and Y together.
datafile: Data file.
          "rg2d.dat" is a sample 2-d data set (DS1 in SIGMOD paper).

3. Output files:

parafile+schemefile+projfile+datafile-log:
        log information: such as running time, clustering quality
        (as defined in SIGMOD paper)

parafile+schemefile+projfile+datafile-i-cluster:
        clusters based on projection i obtained after phase 3
        (represented as <CF,radius, diameter> )

parafile+schemefile+projfile+datafile-i-refcluster:
        clusters based on projection i obtained after phase 4
        (represented as <CF, radius, dimater>)

Please read sample files:
rg2d.para+rg2d.scheme+rg2d.proj+rg2d.dat-log
rg2d.para+rg2d.scheme+rg2d.proj+rg2d.dat-0-cluster
rg2d.para+rg2d.scheme+rg2d.proj+rg2d.dat-0-refcluster
rg2d.para+rg2d.scheme+rg2d.proj+rg2d.dat-1-cluster
rg2d.para+rg2d.scheme+rg2d.proj+rg2d.dat-1-refcluster
rg2d.para+rg2d.scheme+rg2d.proj+rg2d.dat-2-cluster
rg2d.para+rg2d.scheme+rg2d.proj+rg2d.dat-2-refcluster

With other compiling options (see make_include_rockyroad_gcc2.7.1):

-DLABEL
parafile+schemefile+projfile+datafile-i-label:
        for clustering on projection i, label each data record
        with its belonging cluster.

-DFILTER
parafile+schemefile+projfile+datafile-i-dat-j:
        for clustering on projection i, data records of cluster j

-DSUMMARY
parafile+schemefile+projfile+datafile-i-summary:
        for clustering on projection i, summary data records as
        <n, vector xi, vector sqrt(xi)>