Twenty Newsgroups Classification

Get Mahout

Assume the environment variable $HADOOP_HOME refers to the location where you checked out/installed Hadoop
Assume the environment variable $MAHOUT_HOME refers to the location where you checked out/installed Mahout

After downloading the distribution, unzip/untar it into the directory of your choice and do:

Setup:

In trunk, mvn install // This will compile everything and create the Hadoop Job.
cd examples
If you've run this before, you may want to rm -rf the work and temp directories
ant -f build-deprecated.xml get-files //Note, we are in the process of updating to Maven
mkdir lib //NOTE The next few steps are a workaround for the interim while we fully migrate to Maven
mvn dependency:copy-dependencies -DoutputDirectory=lib
ant -f build-deprecated.xml extract-20news-18828 -Ddest=target

Then, from Hadoop:

emacs $HADOOP_HOME/conf/hadoop-site.xml (add in local settings per quickstart)
$HADOOP_HOME/bin/hadoop namenode -format //Format the HDFS
$HADOOP_HOME/bin/start-all.sh //Start Hadoop
$HADOOP_HOME/bin/hadoop dfs -put $MAHOUT_HOME/examples/work/20news-18828-collapse 20newsInput //Copies the extracted text to HDFS

Bayes

Then, to train the Bayes Classifier using tri-grams:

$HADOOP_HOME/bin/hadoop jar $MAHOUT_HOME/examples/target/mahout-examples-0.1.job org.apache.mahout.classifier.bayes.TrainClassifier -i 20newsInput -o newsmodel -ng 3 -type bayes

This will run 4 map reduce jobs on Hadoop to train the classifier and will take a while on a single node machine. You can monitor the status of these jobs by opening a web browser on your Job Tracker node: http://localhost:50030/jobtracker.jsp

To Test:

Testing is not distributed and is done on the master node of your cluster.
You need to set a large heap size in $HADOOP_HOME/conf/hadoop-env.sh on the master in order to test.

export HADOOP_HEAPSIZE=2500

then,

$HADOOP_HOME/bin/hadoop jar $MAHOUT_HOME/examples/target/mahout-examples-0.1.job org.apache.mahout.classifier.bayes.TestClassifier -p newsmodel -t work/20news-18828-collapse -ng 3 -type bayes

Output might look like:

08/11/07 16:52:39 INFO bayes.TestClassifier: Done loading model: # labels: 20
08/11/07 16:52:39 INFO bayes.TestClassifier: Done generating Model
08/11/07 16:52:57 INFO bayes.TestClassifier: alt.atheism	96.9962453066333	775/799.0
08/11/07 16:53:15 INFO bayes.TestClassifier: comp.graphics	99.28057553956835	966/973.0
08/11/07 16:53:45 INFO bayes.TestClassifier: comp.os.ms-windows.misc	96.95431472081218	955/985.0
08/11/07 16:53:59 INFO bayes.TestClassifier: comp.sys.ibm.pc.hardware	99.59266802443992	978/982.0
08/11/07 16:54:10 INFO bayes.TestClassifier: comp.sys.mac.hardware	99.47970863683663	956/961.0
08/11/07 16:54:28 INFO bayes.TestClassifier: comp.windows.x	99.59183673469387	976/980.0
08/11/07 16:54:38 INFO bayes.TestClassifier: misc.forsale	98.45679012345678	957/972.0
08/11/07 16:54:50 INFO bayes.TestClassifier: rec.autos	99.4949494949495	985/990.0
08/11/07 16:55:04 INFO bayes.TestClassifier: rec.motorcycles	100.0	994/994.0
08/11/07 16:55:16 INFO bayes.TestClassifier: rec.sport.baseball	99.89939637826961	993/994.0
08/11/07 16:55:36 INFO bayes.TestClassifier: rec.sport.hockey	99.89989989989989	998/999.0
08/11/07 16:55:54 INFO bayes.TestClassifier: sci.crypt	99.39455095862765	985/991.0
08/11/07 16:56:05 INFO bayes.TestClassifier: sci.electronics	98.98063200815494	971/981.0
08/11/07 16:56:27 INFO bayes.TestClassifier: sci.med	99.79797979797979	988/990.0
08/11/07 16:56:44 INFO bayes.TestClassifier: sci.space	99.3920972644377	981/987.0
08/11/07 16:57:06 INFO bayes.TestClassifier: soc.religion.christian	99.49849548645938	992/997.0
08/11/07 16:57:24 INFO bayes.TestClassifier: talk.politics.guns	99.45054945054945	905/910.0
08/11/07 16:57:51 INFO bayes.TestClassifier: talk.politics.mideast	98.82978723404256	929/940.0
08/11/07 16:58:13 INFO bayes.TestClassifier: talk.politics.misc	89.93548387096774	697/775.0
08/11/07 16:58:25 INFO bayes.TestClassifier: talk.religion.misc	61.78343949044586	388/628.0
08/11/07 16:58:25 INFO bayes.TestClassifier: =======================================================
Summary
-------------------------------------------------------
Correctly Classified Instances          :      18369	   97.5621%
Incorrectly Classified Instances        :        459	    2.4379%
Total Classified Instances              :      18828

=======================================================
Confusion Matrix
-------------------------------------------------------
a    	b    	c    	d    	e    	f    	g    	h    	i    	j    	k    	l    	m    	n    	o    	p    	q    	r    	s    	t    	<--Classified as
994  	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	 |  994   	a     = rec.motorcycles
0    	976  	0    	0    	0    	0    	0    	0    	0    	0    	1    	0    	0    	0    	0    	0    	0    	0    	2    	1    	 |  980   	b     = comp.windows.x
7    	0    	929  	1    	0    	0    	0    	0    	0    	0    	0    	0    	1    	0    	2    	0    	0    	0    	0    	0    	 |  940   	c     = talk.politics.mideast
0    	0    	0    	905  	0    	0    	1    	0    	0    	0    	0    	0    	0    	0    	0    	0    	3    	0    	1    	0    	 |  910   	d     = talk.politics.guns
4    	1    	4    	27   	388  	1    	0    	1    	0    	5    	1    	1    	2    	2    	149  	7    	2    	33   	0    	0    	 |  628   	e     = talk.religion.misc
3    	0    	0    	0    	0    	985  	0    	1    	0    	0    	0    	0    	0    	1    	0    	0    	0    	0    	0    	0    	 |  990   	f     = rec.autos
0    	0    	0    	0    	0    	0    	993  	1    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	 |  994   	g     = rec.sport.baseball
0    	0    	0    	0    	0    	0    	1    	998  	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	 |  999   	h     = rec.sport.hockey
0    	0    	0    	0    	0    	0    	0    	0    	956  	0    	2    	0    	0    	0    	0    	0    	0    	0    	2    	1    	 |  961   	i     = comp.sys.mac.hardware
0    	0    	0    	0    	0    	0    	0    	0    	0    	981  	0    	0    	5    	0    	0    	1    	0    	0    	0    	0    	 |  987   	j     = sci.space
0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	978  	0    	1    	0    	0    	0    	0    	0    	2    	1    	 |  982   	k     = comp.sys.ibm.pc.hardware
1    	0    	3    	36   	0    	1    	2    	1    	0    	5    	0    	697  	4    	0    	3    	3    	19   	0    	0    	0    	 |  775   	l     = talk.politics.misc
0    	2    	0    	0    	0    	0    	0    	0    	0    	0    	2    	0    	966  	0    	0    	0    	0    	0    	2    	1    	 |  973   	m     = comp.graphics
1    	0    	0    	0    	0    	0    	0    	0    	0    	0    	6    	0    	0    	971  	0    	0    	0    	0    	3    	0    	 |  981   	n     = sci.electronics
1    	0    	0    	0    	0    	0    	0    	0    	1    	0    	0    	0    	0    	0    	992  	1    	0    	1    	0    	1    	 |  997   	o     = soc.religion.christian
0    	0    	0    	0    	0    	0    	0    	0    	0    	0    	1    	0    	0    	0    	0    	988  	0    	0    	0    	1    	 |  990   	p     = sci.med
0    	0    	0    	2    	0    	0    	0    	0    	0    	0    	0    	0    	2    	1    	0    	0    	985  	0    	1    	0    	 |  991   	q     = sci.crypt
0    	0    	0    	1    	1    	0    	0    	0    	0    	1    	0    	0    	1    	0    	19   	0    	1    	775  	0    	0    	 |  799   	r     = alt.atheism
1    	0    	0    	0    	0    	3    	1    	2    	0    	0    	3    	0    	0    	5    	0    	0    	0    	0    	957  	0    	 |  972   	s     = misc.forsale
0    	0    	0    	8    	0    	0    	0    	0    	0    	0    	6    	0    	6    	0    	0    	0    	0    	0    	10   	955  	 |  985   	t     = comp.os.ms-windows.misc

Complementary Naive Bayes

To Train a CBayes Classifier using bi-grams

$HADOOP_HOME/bin/hadoop jar $MAHOUT_HOME/examples/target/mahout-examples-0.1.job org.apache.mahout.classifier.bayes.TrainClassifier -i 20newsInput -o newsmodel -ng 2 -type cbayes

To Test a CBayes Classifier using bi-grams

$HADOOP_HOME/bin/hadoop jar $MAHOUT_HOME/examples/target/mahout-examples-0.1.job org.apache.mahout.classifier.bayes.TestClassifier -p newsmodel -t work/20news-18828-collapse -ng 2 -type cbayes