[SPARK-20208][R][DOCS] Document R fpGrowth support

[zero323]

What changes were proposed in this pull request?

Document fpGrowth in:

• vignettes
• programming guide
• code example

How was this patch tested?

Manual tests.

[SparkQA]

Test build #75588 has finished for PR 17557 at commit 27e94fd.

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

[SparkQA]

Test build #75593 has started for PR 17557 at commit 30949a1.

[SparkQA]

Test build #75602 has finished for PR 17557 at commit 2f4c70e.

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

[SparkQA]

Test build #75605 has finished for PR 17557 at commit eb4939d.

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

[felixcheung]

thanks! - I'd prefer example with real data...

[zero323]

What do you mean by "real"? Something human readable (e.g. milk, bread, butter) or some standard pattern mining dataset? If the former one then it is not a problem. If the latter one I am not aware of any dataset which would be safe enough on the license side.

[felixcheung]

something that is not coded in 3 lines ;)
reading from a file if we could - if there isn't any dataset that we can license to use, can we anonymize an existing one?

[zero323]

something that is not coded in 3 lines ;)

That's for sure :) For now I am more trying to figure out how to present this to make it useful. For ML guide we can safely reuse data/mllib but I don't think we can do the same with vignette unless we bring sample_fpgrowth.txt as a package data.

[SparkQA]

Test build #75633 has finished for PR 17557 at commit 03afd08.

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

[zero323]

@felixcheung For vignette I used a bit larger synthetic dataset which should show all the features implemented by fpm. For examples I used the same data as #17130.

[felixcheung]

perhaps it's slightly less clear, since there are 3 references to "items" (or really, just the SparkDataFrame and its column name), which "items" L923 is referring to?

[felixcheung]

I like the approach you have there
https://github.com/apache/spark/pull/17557/files#diff-1d0d34d8ea18a9340f0a02c6befe6269R30

[zero323]

@felixcheung Updated.

BTW There is a JIRA tracking SQL functions parity, isn't there?

[felixcheung]

do you mean making sure we have all the SQL functions in R?
we don't, actually, since it's a evolving tasks - there are constantly new functions being added.

I think you are referring to split - yes we should probably add that in R too

[felixcheung]

nit: could you please rename the dataframe to df like the other example you have too?

[zero323]

I was pretty sure I've seen one :) split and array. There are of course name conflicts involved (spark.array?) but it would be really useful to have these.

[felixcheung]

that's possible but I'm fairly certain there are still quite a few functions we have missed over the year that are not in JIRA.

feel free to add them - would appreciate your help.

agree array is a bit tricky - I'd rather not having to diverge because of consistency with array_contain function and so on, but I can see spark.array might be an approach. Or perhaps array_col?

[zero323]

Or maybe create_array (like PySpark create_map)?

[SparkQA]

Test build #75854 has finished for PR 17557 at commit 4c02933.

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

[SparkQA]

Test build #75859 has finished for PR 17557 at commit ab251f1.

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

[felixcheung]

LGTM except one issue

[felixcheung]

oops. missed this - this should be {r}

[SparkQA]

Test build #75893 has finished for PR 17557 at commit 3445815.

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

[felixcheung]

merged to master/2.2

[zero323]

Thanks @felixcheung!