building
--------------------------------------------------------------------------------
compile the project using ant with the target "package" to generate an
executable jar file:
ant package
attribute detection
--------------------------------------------------------------------------------
the first attribute containing class/Class that is of type string or nominal is
used as class attribute. The first attribute containing document/Document that
is of type string is used as document attribute. If no attribute matching
class/Class or document/Document is found the run is aborted.
options
--------------------------------------------------------------------------------
-i index files
optional, multiple occurrences allowed.
the parameter is split on ",", furthermore the wildcards "*" and
"?" are allowed.
if the parameter is omitted the current working directory is
searched for ".arff" and ".arff.gz" files.
-k the k parameter
optional, one occurrence max.
if the parameter is omitted the value 5 is used.
-m the distance measure to be used (either L1 or L2)
optional, one occurrence max.
if the parameter is omitted the value L1 is used.
-q the query (for the bonus exercise)
optional, multiple occurrences allowed.
if the parameter is omitted a normal document query is started.
if the parameter is given the retrieval with the query is started.
QUERIES
if -q is not used all remaining arguments are treated as documents
for retrieval.
for the bank corpus it is advised that the JVM memory limit is increased with
the "-Xmx" paramter (e.g. -Xmx2048M)
run
--------------------------------------------------------------------------------
example call for document retrieval:
java -jar retrieval.jar -i "arff/news_*grams*" -k 10 -m L2 comp.sys.ibm.pc.hardware/60539 soc.religion.christian/21697 soc.religion.christian/21784
example call for document retrieval with query:
java -jar retrieval.jar -q -i "arff/news_*grams*" word1 word2