These are files for the hierarchichal multi-label version of C4.5, as
described in my PhD thesis:

Clare, A. (2003) Machine learning and data mining for yeast functional
genomics. PhD thesis. University of Wales Aberystwyth. 
http://users.aber.ac.uk/afc/papers/AClarePhDThesis.pdf (1Mb)

They should be untarred over the top of Ross Quinlan's C4.5 Release 8,
which can be downloaded from
http://www.cse.unsw.edu.au/~quinlan/c4.5r8.tar.gz

So to install you should do,

tar xvfz c4.5r8.tar.gz
tar xvfz HierMultiLabelC4.5.tar.gz
cd R8/Src
make all

.data and .names files follow normal C4.5 conventions except that we
can allow multiple class labels per data item, and provide a class
hierarchy.  Multiple classes for each data item should be spearated by
the '@' character. For example:

3.4,2.5,sunny,class1@class2.

The class hierarchy is shown by indentation in a separate file with
the same filestem as the .data and .names files, but ending in
.classes (eg foo.data, foo.names, foo.classes). For example:

a
 a1
  a11
 a2
b
 b1
  b12
 b2
  b21
  b22
   b221


Example data is given in testhier.data, testhhier.names and
testhier.classes. 

Windowing and attribute subsets options to c4.5 do not work - I
haven't updated this part of the code. I developed this only to use
c4.5 and c4.5rules - any other executables are untested and used at
your own risk. 

Confusion matrices are still produced, but should really be ignored -
how can confusion matrices of multi-label problems be represented
truthfully? If a classification is wrong, which column should it be
reported under?

Email me if any problems: afc@aber.ac.uk