Pandas to_datetime 没有错误格式错误

Question

I read in a CSV file containing dates. Some dates may be formatted wrong and I want to find those. With the following approach I would expect the 2nd row to be NaT . But pandas seems to ignore the specified format no matter if I set infer_datetime_format or exact .

import pandas as pd
from io import StringIO

DATA = StringIO("""date
2019 10 07
   2018 10
""")
df = pd.read_csv(DATA)

df['date'] = pd.to_datetime(df['date'], format="%Y %m %d", errors='coerce', exact=True)

results in

        date
0 2019-10-07
1 2018-10-01

The pandas.to_datetime documentation refers to strftime() and strptime() Behavior but when I test it with plain Python it works:

datetime.datetime.strptime('  2018 10', '%Y %m %d')

I get the expected value error:

ValueError: time data '  2018 10' does not match format '%Y %m %d'

What do I miss?

FYI: This question pandas to_datetime not working seems to be related but is different and it seems to be fixed by now. It is working with my pandas version 0.25.2.

Answer 1

This is a known bug, see github for details.

Since we needed a solution I came up with the following workaround. Please note that in my question I used read_csv to keep the reproducible code snippet small and simple. We actually use read_fwf and here is some sample data (time.txt):

2019 10 07 + 14:45 15:00  # Foo
2019 10 07 + 18:00 18:30  # Bar
  2019 10 09 + 13:00 13:45  # Wrong indentation

I felt stating the row number is also a good idea so I added a little bit more voodoo:

class FileSanitizer(io.TextIOBase):
    row = 0
    date_range = None

    def __init__(self, iterable, date_range):
        self.iterable = iterable
        self.date_range = date_range

    def readline(self):
        result = next(self.iterable)
        self.row += 1
        try:
            datetime.datetime.strptime(result[self.date_range[0]:self.date_range[1]], "%Y %m %d")
        except ValueError as excep:
            raise ValueError(f'row: {self.row} => {str(excep)}') from ValueError
        return result


filepath = 'time.txt'
colspecs = [[0, 10], [13, 18], [19, 25], [26, None]]
names = ['date', 'start', 'end', 'description']

with open(filepath, 'r') as file:
    df = pd.read_fwf(FileSanitizer(file, colspecs[0]),
                     colspecs=colspecs,
                     names=names,
                     )

The solution is based on this answer How to skip blank lines with read_fwf in pandas? . Please note this will not work with read_csv .

Now I get the following error as expected:

ValueError: row: 3 => time data '  2019 10 ' does not match format '%Y %m %d'

If anyone has a more sophisticated answer I'm happy to learn.

Answer 2

There is an issue discussing this same aspect of pd.to_datetime with regards to exact string matching.

The thing is that if format is specified and exact is set to True , its a.match like search, meaning it must match at the beginning (as opposed to anywhere). So even though a given date is missing a day, it is a valid match.