[
https://issues.apache.org/jira/browse/MAHOUT-211?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12784599#action_12784599
]
Jake Mannix commented on MAHOUT-211:
------------------------------------
I guess some thought should also be put into how much should be done to our
current implementations, and how much should be done on MAHOUT-204 in the way
of replacing our implementations with colt impls.
> DenseMatrix.getRow() and getColumn() return deep copies of the data
> -------------------------------------------------------------------
>
> Key: MAHOUT-211
> URL: https://issues.apache.org/jira/browse/MAHOUT-211
> Project: Mahout
> Issue Type: Bug
> Components: Matrix
> Affects Versions: 0.1
> Environment: all
> Reporter: Jake Mannix
> Priority: Minor
> Fix For: 0.3
>
>
> As mentioned in the summary, DenseMatrix.getRow() returns a new
> DenseVector(values[row]) instead of just a shallow reference. This is a)
> wasteful, and b) inconsistent with what SparseRowMatrix / SparseColumnMatrix
> do, which means that users who think they can mutate rows by calling getRow()
> are wrong for DenseMatrix, or conversely think they can modify the row at
> will are wrong for SparseRowMatrix.
> Either way, we need to be consistent. I would suggest the contract be that
> you get a shallow view reference (meaning, it may be an actual reference to a
> real Vector, or it may be a virtual Vector, dynamically constructed, which is
> still backed by entries in the matrix), so that calling
> Matrix.getRow(row).set(col, value) has the same affect as Matrix.set(row,
> col, value). If we don't do it this way, we should at least provide an api
> to get a shallow reference to rows.
> Thoughts?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
