inplace dropna?

[jseabold]

see related discussion in #2325

Possible and/or desirable to add an inplace keyword to dropna? Maybe an out keyword, so you can avoid typing the assignment? [Edit: I guess out doesn't really make sense like it does for numpy ufuncs.] Don't really mind if there's a copy even I just get tired of always doing assignment to the same name when I'm doing pre-, post-processing, etc.

dataframe = dataframe.action()

[wesm]

What do you think about having methods like idropna for shorter in-place modification?

[changhiskhan]

doesn't the i- prefix usually indicate positional indexing? like irow, icol, and get_value vs iget_value

[wesm]

I was thinking re: __iadd__ which is the inplace add magic method

[jseabold]

Well, now I'd associate it with iadd, but my first reaction was with indexing as well. I don't really know what to do here. It doesn't kill me to type inplace=True. I'm in the habit now and that's how I noticed dropna didn't have it. If you wanted to be cute (I don't necessarily)

dataframe.ip.dropna()
dataframe.ip.rename(...)

[jtratner]

Now that we have a different representation for NDFrame, couldn't this be as simple as doing something along the lines of:

1. Do dropna or drop.
2. Set index and _data to that of the created object.
3. Delete the other object.
4. delete/clear all of the caches on the original object (both _cache for cache_readonly and the other cacher things.

Don't return self if inplace.

That way references to the original NDFrame wrapper refer to the new object. Easy to do with Series too now that it doesn't inherit NDFrame.

[jtratner]

and then call finalize to copy over any other metadata.

We could rely on __finalize__ to handle tseries/other attributes intelligently.