mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From DAN HELM <>
Subject Re: Questions on RowId job
Date Mon, 04 Jun 2012 11:20:39 GMT
Hmmm. Unless I misunderstand the question, I have always called rowid after having generated
sparse vectors via seq2sparse with the namedVector flag, and it seems to always generate output e.g.,
 $MAHOUT seq2sparse \
    -i ${WORK_DIR}/reuters-out-seqdir/ \
    -o ${WORK_DIR}/reuters-out-seqdir-sparse-cvb \
    -wt tf -seq -nr 3 --namedVector \
  $MAHOUT rowid \
    -i ${WORK_DIR}/reuters-out-seqdir-sparse-cvb/tf-vectors \
    -o ${WORK_DIR}/sparse-vectors-cvb \
I also developed a quick and dirty rowid mod so that it generates one martrix output file
for each "part" input file from seq2sparse command -- so multiple mappers are used to process
output from rowid (such as when running CVB clustering on rowid output).  For that I just
made a derivative rowid command called rowiddist.  So the final solution should be logic
rolled into a single rowid command, where the run type (multiple file output vs single matrix
file) is driven by parameters.  The code below was just a quick "solution".
**** code follows ****
package home.mahout;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.util.ToolRunner;
import org.apache.mahout.common.AbstractJob;
import org.apache.mahout.common.Pair;
import org.apache.mahout.common.iterator.sequencefile.PathFilters;
import org.apache.mahout.common.iterator.sequencefile.PathType;
import org.apache.mahout.common.iterator.sequencefile.SequenceFileDirIterable;
import org.apache.mahout.math.VectorWritable;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.util.Map;
public class RowIdJobDistributed extends AbstractJob {
  private static final Logger log = LoggerFactory.getLogger(RowIdJobDistributed.class);
  public int run(String[] args) throws Exception {
    Map<String,String> parsedArgs = parseArguments(args);
    if (parsedArgs == null) {
      return -1;
    Configuration conf = getConf();
    FileSystem fs = FileSystem.get(conf);
    Path outputPath = getOutputPath();
    Path indexPath = new Path(outputPath, "docIndex");
    SequenceFile.Writer indexWriter = SequenceFile.createWriter(fs,
    FileStatus fstatus[] = fs.listStatus(getInputPath()); 
    try {
      IntWritable docId = new IntWritable();
      int i = 0;
      int numCols = 0;
      int matrixCount = 0;
      for ( FileStatus f: fstatus ) { 
       if (f.isDir()) {
     Path inPath = new Path(f.getPath().toUri().toString());
     Path matrixPath = new Path(outputPath, "matrix-"+matrixCount);
     SequenceFile.Writer matrixWriter = SequenceFile.createWriter(fs,
        for (Pair<Text,VectorWritable> record :
             new SequenceFileDirIterable<Text,VectorWritable>(inPath,
conf)) {
          VectorWritable value = record.getSecond();
          indexWriter.append(docId, record.getFirst());
          matrixWriter.append(docId, value);
          numCols = value.get().size();
      }"Wrote out matrix with {} rows and {} columns to {}", new Object[] { i,
numCols, outputPath });
      return 0;
    } finally {
  public static void main(String[] args) throws Exception { RowIdJobDistributed(), args);


 From: Suneel Marthi <>
To: "" <> 
Sent: Monday, June 4, 2012 1:39 AM
Subject: Questions on RowId job
I have used the RowId Job several times before (and had even advised people on its use in
these forums), so its kinda embarrassing that I ask this.

I have 'Named Vectors' generated via seq2sparse and am trying to create a matrix from these
vectors. RowId Job presently does not generate a matrix when the input is of type 'NamedVector'. 
The code executes successfully to completion with no output and I would have expected it to
throw an Exception of some kind.  

Looking at the code, its obvious that 'Named Vectors' are not being accounted for while generating
a matrix from the input. 

Would it make sense to submit a patch to get RowId Job working with 'Named Vectors' ?

Here are the issues I see with the RowId Job:-

1. If the input has no vectors, RowId Job presently executes successfully and generates a
0 x 0 matrix. (I would have expected an exception of some sort with an appropriate message
like 'Input vectors not found'). 

2. RowId Job presently cannot handle Named Vectors.

3. There was an issue that was brought up on these forums sometime last week to which Jake's
response was that it would be a welcome patch to be able to configure the RowId Job to run
in a distributed mode with many mappers. I agree that this is a useful change. This change
means that the RowSimilarity Job which takes as inputs, the matrix and the number of columns
generated by the RowId job needs to be tested to ensure that nothing's broken.

Should I go ahead and open a Jira for the above issues?

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message