ENH: Import testing into main namespace.

[jseabold]

Unfortunately I won't be able to use this in production until 2020 when we require 0.14, but this would make some things a bit easier for projects that use pandas as a library.

[jreback]

what r u going to use this for?

[jseabold]

Testing without an extra import and using the make* functions interactively for examples, bug-reporting, etc.

[jseabold]

E.g., we need the pandas test function for testing DataFrame equality.

Incidentally, I started to add to this suite some useful functions (DataFrame approximately equal and allclose) until I found that these functions are undocumented and most are written in Cython. Will have to stick to 3+ function calls to test until I find some time for a PR.

[jreback]

their is a new .equals function which should work for you (handles all the nan vs. nan issue)
(also array_equivalent for arrays); could be modified with a precision i think

http://pandas.pydata.org/pandas-docs/dev/whatsnew.html#api-changes

[jseabold]

For precision issues cf. np.testing.assert_allclose. Would be a good target. I don't know what array_equivalent is.

Hmm, equals is an interesting addition. Still nice to keep testing functions in the test namespace, though right?

[jreback]

equals is not really for testing (though it may be used that way). Its a first class function; comparing frames with nans for equality is not possibly right now (without the expressions equals uses); and object arrays are a disaster. see the examples.

why do you want to import testing to the main namespace, can't you import into statsmodels ? (if that's where you are using I assume)

[jseabold]

Numpy imports testing into the main namespace, and I got tired of typing two imports for pandas everywhere. Also for bug reporting. Everyone comes up with there own pd.DataFrame(np.random.random.... Would be simple to just point people to pd.util.testing.make* functions. What's the harm? The kitchen sink is already imported in the __init__ and it doesn't add another actual namespace to the main one.

[jreback]

ok...np then....do you want in 0.13.1 (ok with me)...if so go ahead and merge

[jseabold]

Thanks. Doesn't really matter to me. I'm usually on dev version, and like I said I won't ever actually be able to use this in production until our min version is wherever it ends up (which feels like never these days...).

[jreback]

ok...let's wait for 0.14 then

[ghost]

@jseabold , I thought you were an academic. What constitutes "production"? is there
a GMM capture-the-flag colloquium you run or something?

[jreback]

@jseabold don't you want to do (in __init__)

from pandas.utils import testing as testing

as I presume them you want to do

panel = pd.testing.makePanel()

?

[jseabold]

I was leaving it in the util namespace so that the changes are really very minimal, but flat is better than nested, and I'm fine with it if y'all are.

[jreback]

@jseabold this is fine