pandas 将字符串评估为数字

Question

假设 df 为；

data = {'duration':['1week 3day 2hour 4min 23', '2hour 4min 23sec', '2hour 4min', np.nan, '', '23sec']}
df = pd.DataFrame(data)

我正在尝试将持续时间计算为秒数总和。将值替换为：

df['duration'] = df['duration'].str.replace('week', '*604800+') \
                                                                 .str.replace('day', '*604800+') \
                                                                 .str.replace('hour', '*3600+') \
                                                                 .str.replace('min', '*60+') \
                                                                 .str.replace('sec', '')  \
                                                                 .str.replace(' ', '')

但不能运行 eval 函数，如（pd.eval、apply.eval、eval 等）。某些单元格以“+”号或其他 string/na 问题结尾。有帮助吗？

Ps: 这不是一个重复的问题。

Answer 1

您可以将正则表达式与自定义函数结合使用，将周替换为 7 天，并在单独的数字上添加秒数（您可以添加其他单位）。然后转换 to_timedelta:

def change_units(m):
    d = {'week': (7, 'days'), '': (1, 's')}
    _, i, period = m.groups()
    factor, txt = d[period]
    return f'{factor*int(i)}{txt}'
    
df['delta'] = pd.to_timedelta(df['duration'].str.replace(r'((\d)\s*(week|)\b)',
                                                         replace, regex=True))

输出：

                   duration            delta
0  1week 3day 2hour 4min 23 10 days 02:04:23
1          2hour 4min 23sec  0 days 02:04:23
2                2hour 4min  0 days 02:04:00
3                       NaN              NaT
4                                        NaT
5                     23sec  0 days 00:00:23

然后您可以从 TimeDelta 对象中受益，例如转换为 total_seconds:

pd.to_timedelta(df['duration'].str.replace(r'((\d)\s*(week|)\b)',
                                           change_units, regex=True)
               ).dt.total_seconds()

输出：

0    871463.0
1      7463.0
2      7440.0
3         NaN
4         NaN
5        23.0
Name: duration, dtype: float64

Answer 2

我对有不同的方法：

我写了一个函数将字符串转换成秒：

def convert_all(s):
    if not isinstance(s, str):
        # E.g. for np.nan
        return s
    return sum(convert_part(part) for part in s.split())

def  convert_part(part):
    """Convert an individual segment into seconds.

    >>> convert_part('1day')
    86400.0
    """
    if part.isnumeric():
        return float(part)
    in_seconds = {'week': 7*24*60*60, 'day': 24*60*60, 'hour': 60*60, 'min': 60, 'sec': 1}
    for k,v in in_seconds.items():
        if part.endswith(k):
            return float(part.strip(k))*v
    else:
        # Handle error here - just printing for now
        print(part)
        return 0.0

那么你可以使用 Series.apply:

df['duration_sec'] = df['duration'].apply(convert_all)

输出：

                   duration  duration_sec
0  1week 3day 2hour 4min 23      871463.0
1          2hour 4min 23sec        7463.0
2                2hour 4min        7440.0
3                       NaN           NaN
4                                     0.0
5                     23sec          23.0

pandas 将字符串评估为数字

pandas evaluating strings as numeric

python

eval

pandas