Added two wrappers to the gaussian_kde class.

[nicodelpiano]

The wrappers are:

• pdf
• logpdf

[rgommers]

These are new public methods, so should get docstrings.

Adding unit tests is also necessary. Those live in ``scipy/stats/tests/test_kdeoth.py. You can look at existing tests and add one for each new method.

[rgommers]

@nicodelpiano are you planning to finish this?

[nicodelpiano]

Yes, I was too busy with my master, sorry for that.
About the documentation, is it okay to put "Computes the probability density function" for the pdf wrapper (and something similar for logpdf)?

[rgommers]

Probably for pdf the best thing to do is copy the first line of the evaluate docstring (it is a copy after all) and then add one sentence explaining that those two are equivalent and see the evaluate docstring for more details.

And for logpdf something similar.

[nicodelpiano]

I am thinking that the unit test for pdf will be the same as the "test_kde_bandwidth_method" but changing the "evaluate" calls for "pdf".
Do you think this is correct?

[josef-pkt]

For the unit tests, I think the intention of the test would be more obvious if you just compare pdf and evaluate on an example.

for the docstring: does it work to just assign the evaluate docstring,
pdf.__doc__ = evaluate.__doc__ ?

[nicodelpiano]

What do you think of this?

def test_pdf():
    n_basesample = 50
    xn = np.random.randn(n_basesample)

    gkde = stats.gaussian_kde(xn)

    xs = np.linspace(-15,12,25)
    pdf = gkde.evaluate(xs)
    pdf2 = gkde.pdf(xs)
    assert_almost_equal(pdf, pdf2)

[josef-pkt]

Yes I think it's good, it's directly to the point of checking an alias.
Two points:

• add a np.random.seed(1) so we get deterministic repeatable results
• reduce the tolerance in the assert, once you have a seed you can look at how close the values are. They won't be exactly the same because of numerical imprecision, but we could have for example decimal=12, or set atol and rtol for assert_allclose.

[nicodelpiano]

How about this?

def test_pdf():
    np.random.seed(1)
    n_basesample = 50
    xn = np.random.randn(n_basesample)

    # Default
    gkde = stats.gaussian_kde(xn)

    xs = np.linspace(-15,12,25)
    pdf = gkde.evaluate(xs)
    pdf2 = gkde.pdf(xs)
    assert_almost_equal(pdf, pdf2, decimal=12)

[josef-pkt]

looks good to me

I think you can also add the assert_almost_equal(gkde.logpdf(xs), np.log(pdf), ...) to this test function.

[coveralls]

Changes Unknown when pulling 161f1a6 on nicodelpiano:master into * on scipy:master*.

[nicodelpiano]

Is there something wrong in the patch? please let me know!

[josef-pkt]

looks good to me

(I would still prefer full docstrings)

[rgommers]

@nicodelpiano indeed nothing wrong with this patch. Can be merged, with maybe a tweak to the docstrings.

[rgommers]

Merged with some changes in c0864c9. Thanks @nicodelpiano