[SPARK-11468][SPARKR] add stddev/variance agg functions for Column

[felixcheung]

Checked names, none of them should conflict with anything in base

@shivaram @davies @rxin

[SparkQA]

Test build #45109 has finished for PR 9489 at commit 58d0fd9.

• This patch fails SparkR unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #45148 has finished for PR 9489 at commit 7a2a904.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[shivaram]

cc @sun-rui

[davies]

LGTM

[mengxr]

Should we add aliases var and sd in SparkR, which are more familiar to R users?

[felixcheung]

@mengxr possibly.. though there are usage difference with SparkR DataFrame/Column as compared to R data.frame (eg. agg vs aggregate, var(faithful$eruptions) would have been a Column with DataFrame)
It might be more confusing.

[shivaram]

Looks like dply has aggregate functions for variance, standard deviations ? Can we just match the syntax there ? https://www.rstudio.com/wp-content/uploads/2015/02/data-wrangling-cheatsheet.pdf

[felixcheung]

This does work

> head(agg(df, stddev(df$age)))    # same as head(summarize(df, stddev(df$age)))
  stddev_samp(age)
1         7.778175

Also these work

> gd <- groupBy(df, "name")
> head(agg(gd, stddev(df$age)))    # same as head(agg(gd, age = "stddev"))
     name stddev_samp(age)
1    Andy                0
2 Michael               NA
3  Justin                0

I think they match dplyr, so I could name them sd and var, but just as I mention, doesn't work as one would expect on data.frame column vector.

[sun-rui]

@shivaram, I looked at dplyr package, there is no "sd", "var" defined in the package, seems the page is referencing "sd" "var" in R base.

Yes, stddev/variance agg functions we are adding have similar semantic as "sd" and "var" in R, as you can think of a Column as a vector of same type. So we'd better add aliases for them.

Note that var() in R computes sample variance (I think "sd" also uses sample variance, but not verified) instead of population variance.

[sun-rui]

"stddev_pop" before "stddev_samp",

[shivaram]

Yeah in that case lets add aliases for sd and var as well and match the population/sample behavior.

@felixcheung I think similar functions like mean don't match base R data frame but they are useful when used inside an expression and so its fine to have aliases for this.

[SparkQA]

Test build #45581 has finished for PR 9489 at commit 9243fa4.

• This patch fails some tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #45585 has finished for PR 9489 at commit 7498e39.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[sun-rui]

LGTM

[davies]

Merging this into master and 1.6 branch, thanks!