[SPARK-7151] Correlation methods for DataFrame

• Type: Sub-task
• Status: Closed
• Priority: Minor
• Resolution: Duplicate
• Fix Version/s: 1.4.0
• Component/s: ML, SQL
• Labels: dataframe
• Target Version/s: 1.4.0
• Sprint: Spark 1.5 doc/QA sprint
• Assignee: Burak Yavuz
• Reporter: Joseph K. Bradley
• Created: 26/Apr/15 07:01
• Updated: 06/May/15 06:19
• Resolved: 06/May/15 06:19

Description

We should support computing correlations between columns in DataFrames with a simple API.

This could be a DataFrame feature:

myDataFrame.corr("col1", "col2")
// or
myDataFrame.corr("col1", "col2", "pearson") // specify correlation type

Or it could be an MLlib feature:

Statistics.corr(myDataFrame("col1"), myDataFrame("col2"))
// or
Statistics.corr(myDataFrame, "col1", "col2")

(The first Statistics.corr option is more flexible, but it could cause trouble if a user tries to pass in 2 unzippable DataFrame columns.)

Note: R follow the latter setup. I'm OK with either.

Issue Links

• duplicates: SPARK-7239 Statistic functions for DataFrames