DEPR/DOC: rpy deprecation

[jreback]

so we should put up a deprecation notice on rpyimports that in the future the user should use rpy2 directly rather than the less-supported pandas.rpy converters that we can take action say in 0.17.0

discussed extensivley in #7385

xref #7385, #9187 #7309, #8255

• Update the docs with some example of how to port your pandas examples to the rpy2 API -> done with DOC/DEPR: port pandas.rpy to rpy2 guide #10075
• Go over the functionality in pandas and check if some things should be ported to rpy2 (what still works better in pandas?)
  • Handling of NaN values (comment below: DEPR/DOC: rpy deprecation #9602 (comment))
  • index is not preserved in ri2py -> this should be handled in https://bitbucket.org/rpy2/rpy2/issues/285
• The closed PRs -> push them to rpy2?

[jorisvandenbossche]

I put some todo's above.

[sebp]

Regarding the second item in the list, rpy2 currently does not handle missing values correctly, a NaN value will remain NaN in R and not be considered a missing value (NA). The function pandas.rpy.common.convert_to_r_dataframe performed the conversion correctly.

[jorisvandenbossche]

@jreback I moved this to 0.16.1 milestone: I think we should at least have a link the deprecation warning to the docs, where a minimal link is how to port your code. This minimal thing should not be much work, and I will do it this week.

[davclark]

Thanks for highlighting this Aurélien. Is there an issue on the rpy2 issue
tracker for this?
On Jun 16, 2015 10:18 AM, "Aurélien Mazurie, Ph.D." <
[email protected]> wrote:

This has been mentioned above already, but the new recommended conversion
method rpy2.robjects.conversion.py2ri (and
rpy2.robjects.pandas2ri.py2ri_pandasdataframe) doesn't handle NaN
correctly. This is a significant problem for which I haven't been able to
find a solution yet. Is there a way to fix this?

Reply to this email directly or view it on GitHub
#9602 (comment).

[datnamer]

+1

[ajmazurie]

Not that I know of; or my Google-fu failed me.
Aurelien

[jorisvandenbossche]

I opened an issue at rpy2 to track this there as well: https://bitbucket.org/rpy2/rpy2/issue/282/missing-features-in-rpy2s-pandas2ri

The solution is to improve rpy2's conversion functions (in fixing them, or porting some of the pandas code to rpy2 if possible), and contributions for this would be very welcome! (but rest assured, the functions in pandas, although deprecated, will not be effectively removed until these issues are solved)

[lgautier]

@jorisvandenbossche The issue in the link (issue 282 on rpy2's issue tracker on bitbucket) has been closed. rpy2 conversion feature set for pandas data frames is arguably ahead ;-).

[datnamer]

@lgautier Great! thanks for all your hard work on this and other Rpy2 stuff :)

[jreback]

@jorisvandenbossche did we remove this?

[jorisvandenbossche]

No, we did not yet remove this, because there are (or were, should be checked) still some issues with the rpy2 alternative that were better supported in the pandas version. Needs someone to go over this (but ideally someone who uses pandas.rpy / rpy2).

[lgautier]

@jorisvandenbossche Unless the issues are reported in the rpy2 tracker there is a good chance noone will work on them: https://bitbucket.org/rpy2/rpy2/issues?status=new&status=open . If you add a link to an entry in the pandas tracker I can comment back when the issue status is changing (we worked that way for the rmagic before it was moved to rpy2).

[wesm]

It's my preference to deprecate pandas.rpy and remove in pandas 1.0. I think this can be handled by a 3rd-party package

[jorisvandenbossche]

@wesm It's already deprecated for a couple of releases, and we indeed should certainly remove it. At the time of deprecation, there were still some issues in rpy2 (things that were handled better in the pandas.rpy code), but as @lgautier notes above, those should be solved nowadays.
I just want to go through those reported issues and check that they are indeed solved / update our migration docs, and then we can remove the module in the next release.

[jorisvandenbossche]

Was looking into this today, but ran into some issues converting simple dataframes, reported here: https://bitbucket.org/rpy2/rpy2/issues/394/pandas2ripy2ri-on-basic-dataframe-falls

But in any case, will put up a PR to remove the module.

[davclark]

Being gnomish - the above issue has resolved to a documentation issue.