How many data points are you clustering? How many dimensions? On Mon, Nov 26, 2012 at 2:33 PM, Eduard Gamonal <eduard.gamo...@gmail.com>wrote:
> Hi, > I'm doing a MSc at Northeastern and I'm working on analyzing some US > election polls with kmeans. > I'm a beginner with both Mahout and Hadoop. I've been reading the docs > but I'd still appreciate some orientation on these questions: > > * I can transform my input data into vectors and run k-means using the > command line [1]. I downloaded hadoop (1.0.4, working in a real > cluster) and I wrote a program for it. Then I downloaded Mahout and I > saw that there is a jar file included (0.20, single node: > M2_REPO/org/apache/hadoop/hadoop-core/ > 0.20.204.0/hadoop-core-0.20.204.0.jar > ). If I point HADOOP_HOME to my hadoop installation, will mahout use > it? I set HADOOP_HOME in hadoop/conf/hadoop-env.sh, though. > > * I might need to remove some columns of my data set. With Hadoop I > could write a program to tokenize the input and create the data > structures I needed, and then call kmeansdriver. I can use bash to > remove the columns and mahout from command line. should I write a > program instead? > > * How do I write a program for Mahout 0.7 (and Hadoop 1.x), from scratch? > I need to transform the dataset: Vectors should be created only with > the features I want k-means to consider to cluster my data. Then I can > call kmeansdriver. I think I can do both using the explanation of > http://www.odbms.org/download/TamingTextCH06.pdf > > Should the main class extend any other? > How do I deploy it in a cluster with hadoop? > > * it is my understanding that mahout is a framework. I read the code > example in org.apache.mahout.clustering.syntheticcontrol.kmeans. It > extends AbstractJob. I made a new project in Eclipse and copied the > example. My goal was to run it. I tried "java -jar myjar.jar" and > passing my new jar as a parameter to hadoop. What's the correct way of > running a program for mahout? > > Thanks > > [1] https://cwiki.apache.org/MAHOUT/k-means-commandline.html >
