[SPARK-20849][DOC][SPARKR] Document R DecisionTree

[zhengruifeng]

What changes were proposed in this pull request?

1, add an example for sparkr decisionTree
2, document it in user guide

How was this patch tested?

local submit

[felixcheung]

I'd say try to use a data set without . in column name if you can.
Probably would be confusion when examples are causing warnings when users run them

[felixcheung]

actually, I think there's a confusion - I don't mean to change not to use . in formula
I mean the reason why we have warning=FALSE here is because createDataFrame(longley) will cause a warning because it has column with name with . in it. And we should avoid that if we can

[SparkQA]

Test build #77226 has finished for PR 18067 at commit f43ebe0.

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

[SparkQA]

Test build #77230 has started for PR 18067 at commit 65cf494.

[zhengruifeng]

Jenkins, retests this please

[zhengruifeng]

Jenkins, retest this please

[SparkQA]

Test build #77239 has finished for PR 18067 at commit 65cf494.

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

[zhengruifeng]

@felixcheung Updated. By the way, I update other formulas in sparkr-vignettes.Rmd.

[felixcheung]

ok, then could you check if we could remove
, warning=FALSE?

[felixcheung]

actually, I think there's a confusion - I don't mean to change not to use . in formula
I mean the reason why we have warning=FALSE here is because createDataFrame(longley) will cause a warning because it has column with name with . in it. And we should avoid that if we can

[SparkQA]

Test build #77325 has finished for PR 18067 at commit f413081.

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

[SparkQA]

Test build #77324 has finished for PR 18067 at commit 76a9726.

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

[felixcheung]

as commented, before, please check. I'm pretty sure createDataFrame(longley) will cause a warning

longley
     GNP.deflator     GNP Unemployed Armed.Forces Population Year Employed
1947         83.0 234.289      235.6        159.0    107.608 1947   60.323
1948         88.5 259.426      232.5        145.6    108.632 1948   61.122

so our options are:

• don't use longley (my earlier suggestion)
• use longley but keep warning=FALSE

[zhengruifeng]

option 2: do you mean using ````{r, warning=FALSE}` like other examples?
I think both are OK,.
which do you prefer?

[felixcheung]

yes - but as mentioned, if you can think of a data set that doesn't have dot in column name, like as.data.frame(Titanic)

[SparkQA]

Test build #77349 has finished for PR 18067 at commit 48a9686.

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

[SparkQA]

Test build #77351 has finished for PR 18067 at commit 1a97e42.

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

[felixcheung]

why change to classification for trees?

[zhengruifeng]

@felixcheung just because dataset Titanic is often used to illustrate classification. Like the usage in Kaggle contest[https://www.kaggle.com/c/titanic#evaluation].

[felixcheung]

merged to master