ENH/CLN: Make df.pivot to the mapping of pivot_table

[sinhrks]

Related to #6913.

DataFrame.pivot function uses separate logic and has less keyword options than pandas.tools.pivot_table. How about making DataFrame.pivot to call pivot_table internally to support same keywords as pivot_table?

[jreback]

I remember an issue about this, but can't find it now. What API change would this entail, e.g. visible to user?

[sinhrks]

From user, keywords only in pivot_table (aggfunc, fill_value, margins and dropna) will be available in DataFrame.pivot

[jreback]

oh...this is just a pass-trhu of keywords....ok then

[sinhrks]

Thanks, let me try.

[jorisvandenbossche]

I don't know if it is that simple: there is also a pandas.pivot function, which is what DataFrame.pivot calls, and this is clearly distinct from pandas.pivot_table. So I think the question is also: is there a reason there are two functions (pivot/pivot_table)? Have they incompatible functionality? Or could we maybe merge them into one?

Because it would be strange to have a pd.pivot but a DataFrame.pivot calling pd.pivot_table. So I think we should think it a bit more through.

[sinhrks]

I've compared pivot and pivot_table, and understand that these are different function.

• pivot only intends pivotting, not aggregation. Thus it can handle non-numeric values, such as string.
• pivot_table intends pivotting with aggregation.

I noticed that DataFrame.pivot_table methods already exists separatelly, but not listed in api.rst (briefly explained in reshape.rst). So now I understand:

• Both functions should not be changed.
• Include DataFrame.pivot_table to api.rst.

If it is OK, I'll prepare.

[jorisvandenbossche]

More complete docs are certainly good! Maybe adding somewhere a better explanation of the difference between both functions could also be helpfull (in reshaping.rst or in the docstring of both (see also section)).

[nygeog]

Hi (I'm new) but came into a problem with the 0.13.1 pivot_table function.

Not totally sure if this is relevant to this discussion... but I lifted my question from http://stackoverflow.com/questions/23366016/pandas-0-13-1-pivot-table-pivot-error-not-in-0-12-0-should-and-how-i-go

So I have two lines of code below.

One that works successfully with Python 2.7.5 and Pandas 0.12.0

dfp = pd.pivot_table(df, values=pivYears, rows=['partid'],cols=['finestflag_OV'], aggfunc=np.sum)
And this one that does not work successfully with Python 2.7.6 and Pandas 0.13.1. I'm getting error refering to

malloc: *** mach_vm_map …. error code = 3
here's the code:

dfp = pd.pivot_table(df, values=pivYears, index=['partid'],columns=['finestflag_OV'], aggfunc=np.sum)
It could be that I am misunderstanding pivot vs. pivot_table and/or how it works in 0.13.1.

Lastly, since I've been using 0.12.0 for a while and am comfortable with it, is it possible to downgrade from 0.13.1 to 0.12.0 on my new machine? (Also, can I go back in Python versions so all my machines have the same version of Python and Pandas?)

Edits for questions: Both OS's are OS X 10.9.2.

Here's a subset of 9 rows of data in the CSV. https://github.com/nygeog/python/blob/master/pandas/pivot_sample.csv

The actual file has ~1.7 million rows. Which works fine on machine with 0.12.0.

Thanks, Danny