ENH: Add bw_method and ind kwargs to kind='kde' plot method

[khgitting]

It'd be really cool when plotting kde's to be able to set the bandwidth used for the kde and the indices at which it is evaluated. This is done with the bw_method kwarg for scipy.stats.gaussian_kde for scipy >= 0.11, a couple years old now, for the former; a variable ind is created in the pandas class KdePlot's _make_plot method and passed to gkde.evaluate.

Here's a patch which implements this functionality: khgitting/pandas@master...kde_kwargs

I'm having trouble passing the Travis CI tests here: https://travis-ci.org/khgitting/pandas/jobs/9284591

It appears the tests use an older version of scipy where gaussian_kde does not take bw_method kwarg. I'd love to be able to resolve this, but am not sure what the right course of action is. Any advice?

[jtratner]

Given that pandas has to support that version of scipy, you could check for version and go from there, but you need to figure out a version number that is guaranteed to have the argument for bw_method.

As an aside self.ind is None is preferred to self.ind == None.

[jreback]

see here for how to test for versions: https://github.com/pydata/pandas/wiki/Tips-&-Tricks

[jtratner]

to be very explicit:

from distutils.version import LooseVersion
kwargs = dict()
if LooseVersion(scipy.__version__) >= SCIPYVERSION:
    kwargs["bw_method"] = self.bw_method
gkde = gaussian_kde(y, **kwargs)

Given that it looks like there is only one thing you are setting, you could skip the dictionary creation and do:

if RIGHT_VERSION:
    gkde = gaussian_kde(y, bw_method=self.bw_method)
else:
    gkde = gaussian_kde(y)

[khgitting]

Thanks much. bw_method was added in scipy 0.11.0. I added the version check, corrected the '== None', also added a warning to alert the user for the case when scipy version < 0.11.0 but when bw_method is specified but ignored, and passed the Travis-CI test.

khgitting#1

[khgitting]

I rebased and squashed to reduce to one commit, then pushed, and began failing Travis-CI. After some work to determine the problem, I tried to start from the beginning, making a new fork, one commit, passed tests locally, pushed, and again failed Travis-CI tests. It appears to be in code unrelated to my patch:

https://travis-ci.org/khgitting/pandas/jobs/9324896
File "/home/travis/virtualenv/python2.7_with_system_site_packages/local/lib/python2.7/site-packages/pandas-0.12.0.dev_0d06729-py2.7-linux-x86_64.egg/pandas/io/tests/test_data.py", line 145, in test_get_components_nasdaq_100
assert 'AAPL' in df.index
AssertionError

Perhaps current master is not stable?

[jreback]

this is an unrelelated (recent) failing test

[jreback]

check back periodically
once master is fixed, rebase and it will retest

[cpcloud]

@khgitting fixed, rebase and force push should pass travis now

[khgitting]

@cpcloud still failing at same point https://travis-ci.org/khgitting/pandas/jobs/9327972

[cpcloud]

the parent commit for the change u just posted is 54243c7 which is not current master

current master is 424e187

[cpcloud]

see the wiki for git tips

short version, do:

git remote add upstream git://github.com/pydata/pandas.git
git fetch upstream
git rebase upstream/master

barring any merge conflicts, this should replay your commits on top of the latest commits in pandas git master

[khgitting]

@cpcloud D'oh; thanks.

Pull request passing Travis-CI tests and replayed on top of latest pandas git master:
khgitting#1

[khgitting]

One final question on this -- just want to make sure I'm observing proper etiquette. I guess because it passed the Travis-CI test I have a "This pull request can be automatically merged." Should I click to do this? From http://pandas.pydata.org/developers.html , "This request then appears to the repository maintainers, and they will review the code. "

[cpcloud]

@khgitting see my comment on your pr