[jira] Commented: (MAHOUT-130) Vector should allow for other normalize powers than the L-2 norm

["Ted Dunning (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MAHOUT-130?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12717394#action_12717394 ] 

Ted Dunning commented on MAHOUT-130:
------------------------------------


I would recommend externalizing a norm function.  It would also be useful for the norm to support L_0 (number of non-zero elements) and L_infty (max element) as specially cases in addition to L_1 and L_2.

Also, there is no need to allocate a vector here:

+      Vector vec = like();
+      for (int i = 0; i< cardinality(); i++){
+        vec.setQuick(i, Math.pow(getQuick(i), power));
+      }
+      val = vec.zSum();

Better to just keep the sum in a local variable:

      val = 0;
      for (int i = 0; i< cardinality(); i++){
        val += Math.pow(getQuick(i), power);
      }
      val = vec.zSum();

This saves the allocation and improves cache coherence.  Accuracy could conceivably be a few ULPS worse if zSum is implemented really well (which it isn't).


> Vector should allow for other normalize powers than the L-2 norm
> ----------------------------------------------------------------
>
>                 Key: MAHOUT-130
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-130
>             Project: Mahout
>          Issue Type: Improvement
>            Reporter: Grant Ingersoll
>            Assignee: Grant Ingersoll
>            Priority: Minor
>         Attachments: MAHOUT-130.patch
>
>
> Modify Vector to allow other normalize functions for the Vector
> See http://www.lucidimagination.com/search/document/bf3a7a7a004d4191/norm_calculations

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.