ENH: pprint index/columns?

[jseabold]

I haven't looked at the source yet, but do columns ever get pretty printed? It looks like indices do if it's a DatetimeIndex, but not otherwise.

I often find myself doing

df.columns.tolist()

otherwise it's not really readable.

[jreback]

IIRC display.max_seq_items controls this....

@cpcloud ?

[jseabold]

It looks to me like it just truncates the columns with a .... I was thinking more of one column name per row or something that's easier to parse with the eyes.

[jreback]

can you post an example with how you think it should look?

[jseabold]

List printing usually works for me. Even the nice lined up columns of tab completion are easier to parse though.

[~/]                                                                            
[3]: df = pd.DataFrame(np.random.rand(20,4), columns=['A really long name', 'Another really long name', 'A third really long name', 'The last column name'])

[~/]
[4]: df.columns
[4]: Index([u'A really long name', u'Another really long name', u'A third really long name', u'The last column name'], dtype='object')

[~/]
[5]: df.columns.to<TAB>
df.columns.to_datetime      df.columns.tofile
df.columns.to_native_types  df.columns.tolist
df.columns.to_series        df.columns.tostring

[~/]
[5]: df.columns.tolist()
[5]: 
['A really long name',
'Another really long name',
'A third really long name',
'The last column name']

[jankatins]

Please, no linebreaks after each element: I usually use the output to copy it and use it as a row selection df[["col", "col2"]] and I was quite happy with this (I seem to remember that df.columns printed column names without quotes, so I had to always use list(df.columns))

This is what ipython currently produces:

In [7]: l = ["a", "b", "c"]

In [8]: l
Out[8]: ['a', 'b', 'c']

In [16]: df = pd.DataFrame(np.random.rand(40,8), columns=['A really long name', 'Another really long name', 'A third really long name', 'The last column name', "a","b","c","d"]
)

In [17]: list(df.columns)
Out[17]: 
['A really long name',
 'Another really long name',
 'A third really long name',
 'The last column name',
 'a',
 'b',
 'c',
 'd']

In [18]: df.columns
Out[18]: Index([u'A really long name', u'Another really long name', u'A third really long name', u'The last column name', u'a', u'b', u'c', u'd'], dtype='object')

[jseabold]

The counterargument is that you can always slice .columns, so you really don't need to do any copy and pasting. I'm mainly interested in some solution for readability. It's nearly impossible to see what's actually in the data and where, as I use this day to day. The counterargument to me I guess is just use info. Maybe a column_info property would be nice. I don't think it's worth crowing the namespace for though.

Both sides of this are really arguments for some kind of IDE with a variable browser window pane and tab-completion for selecting variable names... It's still the only thing I miss from using Stata.

[jreback]

This is what ipythons interact can do! (in ipython 2.0)....now to get someone to write it......(for frames)

[jseabold]

Well that's interesting. They'll be qtconsole/notebooks-only I take it? Is this the link? I'm way behind on ipython-dev.

[TomAugspurger]

It may be notebook only. I'm guessing the qt console doesn't handle javascript.

[jreback]

yah..i think notebook only

[jreback]

http://jakevdp.github.io/blog/2013/12/05/static-interactive-widgets/

I don't think it would be very hard to have as default a couple of sliders that show the extent of a frame (and then allow hiding/showing of various columns).

[jreback]

http://nbviewer.ipython.org/gist/rossant/9463955

proof of concept!