[
https://issues.apache.org/jira/browse/MAHOUT-1595?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14061439#comment-14061439
]
ASF GitHub Bot commented on MAHOUT-1595:
----------------------------------------
Github user avati commented on the pull request:
https://github.com/apache/mahout/pull/25#issuecomment-48974295
If you take a row view of a sparse row matrix, this code is not even used.
RandomAccessSparseVector.java has its own NonDefaultIterator() which exploits
sparseness and that gets used in a row view of a sparse matrix. Similarly
DenseVector has its NonDefaultIterator() which works on a dense vector is used
in a row view of a DenseMatrix.
However for "custom" matrix (like H2OBlockMatrix in
https://github.com/avati/mahout/blob/MAHOUT-1500/h2o/src/main/java/org/apache/mahout/h2obindings/H2OBlockMatrix.java)
which is neither inherited from SparseMatrix nor from DenseMatrix, viewRow()
returns a MatrixVectorView. In this case no assumption can be made about
sparseness about the custom matrix, and this default implementation must
fall-back to a dense-like implementation (which guarantees to give correct
results no matter the nature of the matrix)
> iterateNonZero() is broken in MatrixVectorView
> ----------------------------------------------
>
> Key: MAHOUT-1595
> URL: https://issues.apache.org/jira/browse/MAHOUT-1595
> Project: Mahout
> Issue Type: Bug
> Components: Math
> Reporter: Anand Avati
>
> iterateNonZero() in MatrixVectorView is broken and behaves like a normal
> iterator, over all elements.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
