flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alexey Diomin (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (FLINK-4148) incorrect calculation distance in QuadTree
Date Mon, 04 Jul 2016 12:45:11 GMT

     [ https://issues.apache.org/jira/browse/FLINK-4148?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Alexey Diomin updated FLINK-4148:
---------------------------------
    Description: 
https://github.com/apache/flink/blob/master/flink-libraries/flink-ml/src/main/scala/org/apache/flink/ml/nn/QuadTree.scala#L105

Because EuclideanDistanceMetric extends SquaredEuclideanDistanceMetric we always move in first
case and never reach case for math.sqrt(minDist)

correct match first EuclideanDistanceMetric and after it SquaredEuclideanDistanceMetric

p.s. because EuclideanDistanceMetric more compute expensive and stay as default DistanceMetric
it's can cause some performance degradation for KNN on default parameters

  was:
https://github.com/apache/flink/blob/master/flink-libraries/flink-ml/src/main/scala/org/apache/flink/ml/nn/QuadTree.scala#L105

Because EuclideanDistanceMetric extends SquaredEuclideanDistanceMetric we always move in first
case and never reach case for math.sqrt(minDist)

correct match first EuclideanDistanceMetric and after it SquaredEuclideanDistanceMetric


> incorrect calculation distance in QuadTree
> ------------------------------------------
>
>                 Key: FLINK-4148
>                 URL: https://issues.apache.org/jira/browse/FLINK-4148
>             Project: Flink
>          Issue Type: Bug
>            Reporter: Alexey Diomin
>            Priority: Trivial
>
> https://github.com/apache/flink/blob/master/flink-libraries/flink-ml/src/main/scala/org/apache/flink/ml/nn/QuadTree.scala#L105
> Because EuclideanDistanceMetric extends SquaredEuclideanDistanceMetric we always move
in first case and never reach case for math.sqrt(minDist)
> correct match first EuclideanDistanceMetric and after it SquaredEuclideanDistanceMetric
> p.s. because EuclideanDistanceMetric more compute expensive and stay as default DistanceMetric
it's can cause some performance degradation for KNN on default parameters



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message