ENH: add the parameter dropna to the function to_dict.

[derDieDasJojo]

this is usefull because you cant reach the same effect by just using
dropna().to_dict() also see
http://stackoverflow.com/questions/26033301/make-pandas-dataframe-to-a-dict-and-dropna/26033302#26033302

[cpcloud]

this needs a test

[jreback]

actually I don't think this should be included - u can put the example in the cookbook of you would like

[derDieDasJojo]

@cpcloud: all right, i can write some tests for it.
@jreback: do you have any arguments for your position ? why do you think, it should not be included ?
I don't know any other elegant way to do this. so i consider it to be a as usefull potential future part of pandas.

[jreback]

• this is easily done as the example from SO
• not sure why this would actually be useful
• in no other places is this done so makes the API inconsistent

[jreback]

maybe if u explain what u r doing with it
why are you using a dict at all

[derDieDasJojo]

why this could be usefull:
I have a dataframe:

>>> data A B 1 2 NaN 2 3 44 3 4 NaN 

and need a dict of that data. because in a dict there is an association by id:value there is no need to list all empty values with id:NaN. Instead typically you just every entry with an value is listed.
So if you want a dict out of a dataframe it is very well possible, that you want something like this:

{'A': {1: 2, 2: 3, 3: 4}, 'B': {2: 44.0}} 

But there is no way to get something like this using pandas methods. So in this case you are back on your own and have to write a dict-comprehension that somewhat like this:

dict((k, v.dropna().to_dict()) for k, v in compat.iteritems(data)) 

Of course it is possible to write your own dict-comprehension every time you need it. But it is easier and more robust if there is already a library doing this for you.
With this enhancement you could take advantage of the already existing to_dict() function for DataFrames and simply write

data.to_dict(dropna=True) 

API consistancy

if you look up DataFrame.to_excel or DataFrame.to_html every one of it has some parameters that are special for this output type and make sence for this one but not for other ones.
And on the same way this could be an optional parameter that makes sence for Data transformation from a DataFrame to a dict.

why using a dict
actually this is no good argument agains a DataFrame.to_dict() function enhancement.
If pandas offers a functionality to reshape data from a DataFrame to a dict, why not using a dict ?
I am using a dict, because another python class has an interface expecting a dict.

Tests
I can write tests for it - thats no problem.
I will do this if this enhancement is considered to be usefull and has a change to be accepted - otherwise they would anyway be useless

[jorisvandenbossche]

Another possibility would be to use the na_rep keyword for this (a keyword used by much of the other to_... functions). Where the default value for this is just the python nan, but you could also provide a string 'NaN' or provide False to not include missing values at all (the dropna from this PR).

I think, if we want to include the functionality proposed in this PR, using dropna is more clear for this specifically, but something like na_rep could have more general use. Of course, having here a possible False for na_rep is something that is not available in the other writers which makes it also a bit inconsistent.

[jreback]

ok will mark as an enhancement

needs tests

[jreback]

closing pls reopen if/when updated

[jorisvandenbossche]

@derDieDasJojo if you would update, I think using a na_rep makes it a bit more generally useful.