BUG: SparseSeries.shift may raise NameError or TypeError

[sinhrks]

• no existing issue
• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

Fixed following bugs on current master. Also, moved TestSparseArrayIndex to test_libsparse

1. shift

pd.SparseSeries([1, 2, 3], fill_value=0).shift(1)
# NameError: global name 'kwds' is not defined

pd.SparseSeries([1, 2, 3]).shift(1)
# TypeError: %d format: a number is required, not float

2. dtype

from pandas.sparse.array import IntIndex
arr = pd.SparseArray([1, 2], sparse_index=IntIndex(4, [1, 2]), dtype=None)
arr.dtype
# dtype('int64')

arr.values
# array([-9223372036854775808,                    1,                    2,
#        -9223372036854775808])

# Expected outputs
arr.dtype
# dtype('float64')

arr.values
# array([ nan,   1.,   2.,  nan])

[jreback]

I would test the fill_value on all of the these (e.g. nan as the default, 0 on next example)

[sinhrks]

Added more tests in test_shift_nan. Is this what you mean?

[jreback]

no I meant that you are asserting the fill_value in some of the tests here (as well as the dtype), but might want to assert fill_value for all (an example will be when we have nan fill values that we DO coerce on creation), e.g. say we have sparse datetimes (we don't really support this ATM)

pd.SparseArray([pd.NaT, pd.Timestamp('20130101'), pd.Timestamp('20130102'), pd.NaT], fill_value=np.nan)

[sinhrks]

@jreback Thanks. ATM we can't create non-numeric SparseArray (#11856). Your example raises TypeError.

[jreback]

no I get it. I just want a line that says self.assertTrue(np.isnan(arr.fill_value)) on the tests that don't have this.

[sinhrks]

I see. Added missing checks.

[jreback]

thanks!

sparse is really looking nice!