[HIVEMALL-78] AUC UDAF for BinaryClassificationMetrics

• Type: New Feature
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 0.5.0
• Assignee: Takuya Kitazawa
• Reporter: Makoto Yui
• Created: 21/Feb/17 10:32
• Updated: 09/Nov/17 03:33
• Resolved: 28/Feb/17 10:19

Description

Support Area Under ROC of Binary Classification Metrics.

-- option 1
select
   auroc(label, prob) as auroc
from
   data;

-- option 2
WITH roc as (
  select
    roc(label, prob) as tpr, tpr
  from
    data
)
select
  auc(fpr, tpr) as auroc -- auc is UDAF, input is sorted by fp asc
from (  
  select 
    fpr, tpr
  from
    roc
  DISTRIBUTE BY 
    floor(fpr / 0.2) -- 5 bins
  SORT BY 
    fpr ASC
) t

Reference)
http://www.citeulike.org/user/myui/article/12615084
https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/evaluation/BinaryClassificationMetrics.scala
https://spark.apache.org/docs/latest/mllib-evaluation-metrics.html

Issue Links

• blocks: HIVEMALL-21 [Umbrella] Release Hivemall v0.5.0
• links to: GitHub Pull Request #52