Inconsistent behaviour in Timestamp.round · Issue #22591 · pandas-dev/pandas (original) (raw)

Code Sample

import pandas as pd a = pd.to_datetime('2000-01-01 0:0:0.124999360') b = pd.to_datetime('2018-01-01 0:0:0.124999360') a.round('us') Timestamp('2000-01-01 00:00:00.124999') b.round('us') Timestamp('2018-01-01 00:00:00.125000')

Problem description

According to the standard definition of rounding to a multiple b.round('us') in the example above gives the wrong answer.

I suspect that there is loss of precision due to a floating point conversion.

>>> round(a.value / 1e9, 6)
946684800.124999
>>> round(b.value / 1e9, 6)
1514764800.125

Expected Output

>>> b.round('us')
Timestamp('2018-01-01 00:00:00.124999')

Output of `pd.show_versions()`

INSTALLED VERSIONS

commit: None
python: 3.7.0.final.0
python-bits: 64
OS: Darwin
OS-release: 17.7.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8