从文本列中提取日期 Pandas

Question

How to extract date from text column: ex:

import pandas as pd

data = [
    [1, "NOV 20/00 I have a date"],
    [2, "DEC 20 I am going to shopping"],
    [3, "I done with all the things"],
    [4, "NOV 10 2021 YES I AM"],
    [5, "JAN/20/2020 - WILL CALL IN DIRECTIONS"],
]

chk = pd.DataFrame(data, columns = ['id', 'strin'])

Required Output:

1 2000-11-20
2 
3 
4 2021-11-10
5 2020-01-20

Answer 1

Not the most pleasant solution, but this 'works'.

I've only added month to month-number mappings for the two months (JAN and NOV) in your sample data, as I can't guess what your other mappings might be. Also no reason you can't add multiple mappings for each month, eg,: 'FBR': 2, 'FEB': 2

Sample data setup:

import pandas as pd

data = [
    [1, "NOV 20/00 I have a date"],
    [2, "DEC 20 I am going to shopping"],
    [3, "I done with all the things"],
    [4, "NOV 10 2021 YES I AM"],
    [5, "JAN/20/2020 - WILL CALL IN DIRECTIONS"],
    [6, "JAN/7/2020 This has single-digit day"]
]

chk = pd.DataFrame(data, columns = ['id', 'text'])

Note I called the 2nd column text as it made more sense to me than strin . Added an extra row to test single-digit day matches.

Code to parse the dates where possible:

import datetime
monthMap = {"JAN" : 1, "NOV" : 11}
monthRegex = "|".join(monthMap.keys())
separatorRegex = "[ /]"
dateRegex = f"^({monthRegex}){separatorRegex}(\d\d?){separatorRegex}(\d\d\d?\d?)"

def mapMatchesToDate(matchRow):
    if matchRow:
        [month, day, year] = matchRow[0]
        month = monthMap[month]
        year = "20" + year if len(year) == 2 else year
        return datetime.date(int(year), int(month), int(day))
    else:
        return None

chk['date'] = chk['text'].str.findall(dateRegex).apply(mapMatchesToDate)

Outputs the pandas Series (ie, just a single column):

0    2000-11-20
1          None
2          None
3    2021-11-10
4    2020-01-20
5    2020-01-07

This could be assigned to a column in the original data with (eg):

chk['date'] = chk['text'].str.findall(dateRegex).apply(mapMatchesToDate)

Meaning that chk is now:

   id                                   text        date
0   1                NOV 20/00 I have a date  2000-11-20
1   2          DEC 20 I am going to shopping        None
2   3             I done with all the things        None
3   4                   NOV 10 2021 YES I AM  2021-11-10
4   5  JAN/20/2020 - WILL CALL IN DIRECTIONS  2020-01-20
5   6   JAN/7/2020 This has single-digit day  2020-01-07

Problems/notes

1. This is not efficient pandas code , and makes very little use of any optimised pandas functionality.
2. In the case that there are any three-digit numbers in the 'year' position, this won't work, eg, NOV 1 123, will make a date in the year 123
3. Any values in the 'year' position with only two digits are assumed to be 2000..., so 17 is 2017, and 99 is 2099 , not 1999
4. Dates are only matched if they are the very first thing in the line. (This is controlled by the ^ character at the start of dateRegex
5. No attempt to match partial dates (eg, DEC 20, in your sample data).
6. Single-digit days (eg "1") are allowed (shown the extra final row in sample data). Single-digit years are not.
7. The only separator characters allowed between date components are space and / . These are defined in separatorRegex