SparseSeries.dropna() returns "NaN"s for values

[rowanu]

A SparseSeries object I have called ".dropna()" on will output the indexes which are not NaN (which is correct), but shows "NaN" in the value column.

If I run SparseSeries.to_dense().dropna() I get the expected output ie. non-NaN indexes and their values.

[wesm]

Need to have a special implementation of dropna for SparseSeries. Need to decide what it should return-- a SparseSeries or a regular dense Series. The behavior might be different depending on what the fill_value is

[rowanu]

My expectation would be a regular dense Series (consisting of only the indexes with values), but I have limited experience in these things.

If someone wanted a SparseSeries returned, why would they be running dropna() over it?

[wesm]

Suppose that you had fill_value=0 but had NAs in the non-zero values?

[wesm]

I fixed this up in git master. Will be part of 0.7.0 final

[rowanu]

Great, thanks for that! What did you end up deciding to do about the return
type?

On Sat, Feb 4, 2012 at 11:07 PM, Wes McKinney <
reply@reply.github.com

wrote:

I fixed this up in git master. Will be part of 0.7.0 final

---

Reply to this email directly or view it on GitHub:
https://github.com/wesm/pandas/issues/730#issuecomment-3815481

[wesm]

When the fill_value is NaN it will return a dense Series. When it's non-NA it will return another SparseSeries. Somewhat "impure" but I think that will be the most useful behavior