[jira] Commented: (MAHOUT-466) simplify or alternative Similarity arithmetic(AbstractDistributedVectorSimilarity) for boolean data

["Sebastian Schelter (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MAHOUT-466?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12897773#action_12897773 ] 

Sebastian Schelter commented on MAHOUT-466:
-------------------------------------------

Could I ask you to provide a patch and some test data that shows the performance increases?

> simplify or alternative  Similarity arithmetic(AbstractDistributedVectorSimilarity) for boolean data
> ----------------------------------------------------------------------------------------------------
>
>                 Key: MAHOUT-466
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-466
>             Project: Mahout
>          Issue Type: Improvement
>          Components: Collaborative Filtering
>    Affects Versions: 0.4
>            Reporter: Hui Wen Han
>             Fix For: 0.4
>
>
> For boolean data ,the prefValue  is  always 1.0f, We need simplify Similarity arithmetic
> for example:
> 1) DistributedEuclideanDistanceVectorSimilarity 
> package org.apache.mahout.math.hadoop.similarity.vector;
> import org.apache.mahout.math.hadoop.similarity.Cooccurrence;
> /**
>     * distributed implementation of euclidean distance as vector similarity measure
>       */
>       public class DistributedEuclideanDistanceVectorSimilarity extends AbstractDistributedVectorSimilarity {
> @Override
> protected double doComputeResult(int rowA, int rowB, Iterable<Cooccurrence> cooccurrences, double weightOfVectorA,
> double weightOfVectorB, int numberOfColumns) {
> double n = 0.0;
> double sumXYdiff2 = 0.0;
> for (Cooccurrence cooccurrence : cooccurrences) { double diff = cooccurrence.getValueA() - cooccurrence.getValueB(); sumXYdiff2 += diff * diff; n++; }
> return n / (1.0 + Math.sqrt(sumXYdiff2));
> }
> }
> this one is always return n (=cooccurrence.size())
> 2) DistributedUncenteredCosineVectorSimilarity 
> /**
>     * distributed implementation of cosine similarity that does not center its data
>       */
>       public class DistributedUncenteredCosineVectorSimilarity extends AbstractDistributedVectorSimilarity {
> @Override
> protected double doComputeResult(int rowA, int rowB, Iterable<Cooccurrence> cooccurrences, double weightOfVectorA,
> double weightOfVectorB, int numberOfColumns) {
> int n = 0;
> double sumXY = 0.0;
> double sumX2 = 0.0;
> double sumY2 = 0.0;
> for (Cooccurrence cooccurrence : cooccurrences) { double x = cooccurrence.getValueA(); double y = cooccurrence.getValueB(); sumXY += x * y; sumX2 += x * x; sumY2 += y * y; n++; }
> if (n == 0) { return Double.NaN; }
> double denominator = Math.sqrt(sumX2) * Math.sqrt(sumY2);
> if (denominator == 0.0) { // One or both vectors has -all- the same values; // can't really say much similarity under this measure return Double.NaN; }
> return sumXY / denominator;
> }
> }
> this one will always return 1.0
> 3) DistributedUncenteredZeroAssumingCosineVectorSimilarity 
> If n users like ItemA, m users like ItemB,p users like both ItemA and ItemB,
> DistributedUncenteredZeroAssumingCosineVectorSimilarity return p/(m*n).
> it also can use for Boolean data, but we can provide a simple one , return (p*p)/(m*n),no so much computing.

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