I've noticed a bunch of issues with SparseDataFrame's constructor, particularly around support for non-float64 datatypes:

• when data is not provided, the default_fill_value is not used to fill the data (noted in passing BUG: SparseDataFrame does not allow single value data #5470):

>>> pd.SparseDataFrame(columns=list("ab"), index=range(4), default_fill_value=0.0) a b 0 NaN NaN 1 NaN NaN 2 NaN NaN 3 NaN NaN

- [ ] when data is not provided and the dtype is set to int64, the constructor call fails

>>> pd.SparseDataFrame(columns=list("ab"), index=range(4), dtype=np.int64) ValueError: cannot convert float NaN to integer

• when data is an empty sparse matrix with int64 entries and no dtype is specified, the type of the SparseDataFrame is still float64. Worse, even when dtype is specified, it still doesn't work.

>>> pd.SparseDataFrame(coo_matrix((4,3), dtype=np.int64)) 0 1 2 0 NaN NaN NaN 1 NaN NaN NaN 2 NaN NaN NaN 3 NaN NaN NaN >>> pd.SparseDataFrame(coo_matrix((4,3), dtype=np.int64), dtype=np.int64) 0 1 2 0 NaN NaN NaN 1 NaN NaN NaN 2 NaN NaN NaN 3 NaN NaN NaN

• probably due to the same issue, given a sparse matrix with underlying elements of type int64, if any of the values in a column is 0 then that column will be treated as float64s:

>>> pd.SparseDataFrame(coo_matrix(np.arange(12).reshape(4,3), dtype=np.int64)) 0 1 2 0 NaN 1 2 1 3.0 4 5 2 6.0 7 8 3 9.0 10 11

(as long as at least one value is non-zero, then this can be worked around by setting default_fill_value; however, if all of the values in one column are zero then this no longer works).

Output of pd.show_versions()