熊猫 read_csv 解析日期

Question

I have written this date parsing function

def date_parser(string):
   try:
       date = pd.datetime.strptime(string, "%d/%m/%Y")
   except:
       date = pd.NaT
   return date

and I call it in pd.read_csv like this

df = pd.read_csv(os.path.join(path, file),
                 sep=";",
                 encoding="latin-1",
                 keep_default_na=False,
                 na_values=na_values,
                 index_col=False,
                 usecols=keep,
                 dtype=dtype,
                 date_parser=date_parser,
                 parse_dates=dates)

The problem is that in one of my dates column, I end up with mixed data types

df[data].apply(type).value_counts()

• class 'datetime.datetime'
• class 'pandas._libs.tslibs.timestamps.Timestamp'
• class 'pandas._libs.tslibs.nattype.NaTType'

I should only have the last two right?

Answer 1

I suggest change your function by to_datetime with errors='coerce' for return NaT if not matched format %d/%m/%Y :

def date_parser(string):
   return pd.to_datetime(string, format="%d/%m/%Y", errors='coerce')