BUG: pandas.cut does not give the right answer with nullable integer Series input

[khdlim]

Code Sample, a copy-pastable example if possible

import numpy
import pandas

input_array = numpy.array([1, 2, numpy.nan, 4, 5])
print('Input array: %s\n' % input_array)

bins = numpy.array([0.5, 2.5, 4.5, 6.5])
print ('Using bins: %s\n' % bins)

test_series = pandas.Series(input_array).astype('Int64')

print('Test series:\n\n%s\n' % test_series)
print('Results of cut:\n\n%s\n'% pandas.cut(test_series, bins))

Console output:

Input array: [ 1.  2. nan  4.  5.]

Using bins: [0.5 2.5 4.5 6.5]

Test series:

0      1
1      2
2    NaN
3      4
4      5
dtype: Int64

Results of cut:

0    (0.5, 2.5]
1    (0.5, 2.5]
2           NaN
3    (0.5, 2.5]
4    (4.5, 6.5]
dtype: category
Categories (3, interval[float64]): [(0.5, 2.5] < (2.5, 4.5] < (4.5, 6.5]]

Problem description

output[3] should be (2.5, 4.5] and not (0.5, 2.5] as shown.
Tested in pandas 1.0.0 and 0.25.3.

Expected Output

Input array: [ 1.  2. nan  4.  5.]

Using bins: [0.5 2.5 4.5 6.5]

Test series:

0      1
1      2
2    NaN
3      4
4      5
dtype: Int64

Results of cut:

0    (0.5, 2.5]
1    (0.5, 2.5]
2           NaN
3    (2.5, 4.5]
4    (4.5, 6.5]
dtype: category
Categories (3, interval[float64]): [(0.5, 2.5] < (2.5, 4.5] < (4.5, 6.5]]

Output of pd.show_versions()

INSTALLED VERSIONS

commit : None
python : 3.6.10.final.0
python-bits : 64
OS : Linux
OS-release : 4.4.0-173-generic
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.0.0
numpy : 1.18.1
pytz : 2019.3
dateutil : 2.6.1
pip : 20.0.2
setuptools : 40.6.3
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.11.1
IPython : 7.11.1
pandas_datareader: None
bs4 : None
bottleneck : None
fastparquet : None
gcsfs : None
lxml.etree : None
matplotlib : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pytables : None
pytest : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
xlsxwriter : None
numba : None

[selasley]

This seems to be a byproduct of converting test_series to object type with NAN (#31290)

bins = np.array([0.5, 2.5, 4.5, 6.5])
x = np.array([1, 2, numpy.nan, 4, 5], dtype='object')
bins.searchsorted(x, side='left')
   array([1, 1, 0, 1, 3])

If x is converted to float64 with NAN instead,

x = np.array([1, 2, numpy.nan, 4, 5], dtype='float64')
bins.searchsorted(x, side='left')
  array([1, 1, 4, 2, 3])

and the result of the call to cut() is

0    (0.5, 2.5]
1    (0.5, 2.5]
2           NaN
3    (2.5, 4.5]
4    (4.5, 6.5]
dtype: category
Categories (3, interval[float64]): [(0.5, 2.5] < (2.5, 4.5] < (4.5, 6.5]]

[TomAugspurger]

This may be fixed on master:

Results of cut:

0    (0.5, 2.5]
1    (0.5, 2.5]
2           NaN
3    (2.5, 4.5]
4    (4.5, 6.5]
dtype: category
Categories (3, interval[float64]): [(0.5, 2.5] < (2.5, 4.5] < (4.5, 6.5]]

[khdlim]

@selasley Thanks for the link, didn't think to search for pd.cut so it didn't show up when i searched for similar issues. Very informative, thanks!

@TomAugspurger Thanks for the heads up for working on this compatibility issue!