BUG: SparseDataFrame indexing sometimes loses fill_value of empty columns in 0.24

[scottgigante]

Code Sample, a copy-pastable example if possible

import numpy as np
import pandas as pd
X = pd.SparseDataFrame([[0,1], [0,0]], default_fill_value=0.0)
## Good behaviour
X.loc[0].to_numpy()
# array([0., 1.])
X.loc[[0]].to_numpy()
# array([[0., 1.]])
X.iloc[0].to_numpy()
# array([0., 1.])
## Bad behaviour
X.iloc[[0]].to_numpy()
# array([[nan, 1]], dtype=object)
X.loc[[True, False]].to_numpy()
# array([[nan, 1]], dtype=object)

Problem description

Indexing a SparseDataFrame with iloc and more than a single row number should return the same result as indexing the same rows with loc and the corresponding indices. Instead, iloc drops column fill_value for any column with no non-zero entries.

Expected Output

All commands should return array([0., 1.]) (allowing for differences between 1- and 2-D output.) The last two (iloc with fancy indexing, and loc with boolean indexing) returns instead array([nan, 1.]).

Output of pd.show_versions()

INSTALLED VERSIONS
------------------
commit: None
python: 3.6.7.final.0
python-bits: 64
OS: Linux
OS-release: 4.4.0-17763-Microsoft
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: C.UTF-8
LOCALE: en_US.UTF-8

pandas: 0.24.1
pytest: None
pip: 18.0
setuptools: 40.2.0
Cython: 0.29
numpy: 1.15.1
scipy: 1.2.0
pyarrow: None
xarray: None
IPython: 7.1.1
sphinx: 1.6.7
patsy: None
dateutil: 2.7.3
pytz: 2018.3
blosc: None
bottleneck: None
tables: 3.4.4
numexpr: 2.6.8
feather: None
matplotlib: 2.2.3
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml.etree: None
bs4: None
html5lib: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.10
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None
gcsfs: None

[TomAugspurger]

FYI we’re likely deprecating SparseDataFrame. You’re probably better off switching to a regular data frame with sparse columns.

…

On Feb 19, 2019, at 5:35 PM, Scott Gigante ***@***.***> wrote: Code Sample, a copy-pastable example if possible import numpy as np import pandas as pd X = pd.SparseDataFrame([[0,1], [0,0]], default_fill_value=0.0) ## Good behaviour X.loc[0].to_numpy() # array([0., 1.]) X.loc[[0]].to_numpy() # array([[0., 1.]]) X.iloc[0].to_numpy() # array([0., 1.]) ## Bad behaviour X.iloc[[0]].to_numpy() # array([[nan, 1]], dtype=object) X.loc[[True, False]].to_numpy() # array([[nan, 1]], dtype=object) Problem description Indexing a SparseDataFrame with iloc and more than a single row number should return the same result as indexing the same rows with loc and the corresponding indices. Instead, iloc drops column fill_value for any column with no non-zero entries. Expected Output All commands should return array([0., 1.]). The last two (iloc with fancy indexing, and loc with boolean indexing) returns instead array([nan, 1.]). Output of pd.show_versions() — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[scottgigante]

I don't see any mention of this in the documentation. Can you please post a link?

[TomAugspurger]

Not deprecated yet: #19239

…

On Wed, Feb 20, 2019 at 12:18 PM Scott Gigante ***@***.***> wrote: I don't see any mention of this in the documentation. Can you please post a link? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#25378 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHImSaByNWKzAsomtNSFDzpvglO6w-ks5vPZGKgaJpZM4bEF0P> .

[gfyoung]

Marking as a bug for now, but given that this isn't a regression, it would likely be patched in 0.25.0. However, even with the deprecation, a patch would be welcomed if it isn't too difficult.