当使用自定义 function 时，Pandas 应用/映射在某些行不起作用

Question

I have the following DataFrame where the CMO _ ($ / MWH) is a finantial value with needs to be rounded using the old rounding method (eg: 754.275 must be rounded to 754.28):

    INIT_DATE  INIT_HOUR   CMO_($/MWH)
0   2020-12-01  00:00:00    754.275
1   2020-12-01  01:00:00    728.130
2   2020-12-01  02:00:00    722.575
3   2020-12-01  03:00:00    722.045
4   2020-12-01  04:00:00    721.950
5   2020-12-01  05:00:00    721.035
6   2020-12-01  06:00:00    722.100
7   2020-12-01  07:00:00    739.925
8   2020-12-01  08:00:00    771.390
9   2020-12-01  09:00:00    797.415
10  2020-12-01  10:00:00    796.585
11  2020-12-01  11:00:00    791.875
12  2020-12-01  12:00:00    782.225
13  2020-12-01  13:00:00    783.540
14  2020-12-01  14:00:00    790.980
15  2020-12-01  15:00:00    815.555
16  2020-12-01  16:00:00    824.760
17  2020-12-01  17:00:00    782.265
18  2020-12-01  18:00:00    779.970
19  2020-12-01  19:00:00    784.640
20  2020-12-01  20:00:00    785.380
21  2020-12-01  21:00:00    785.840
22  2020-12-01  22:00:00    779.775
23  2020-12-01  23:00:00    763.775

Since Python 3+ uses the Bank's Rounding rule, I implemented a custom round function that makes the old rounding method:

def _old_round(n):
    """
        Since Python 3+ the rounding is made through the Banker's Round Rule,
        this function uses the old rounding method (round up if 0.5).
        
        Args:
            n (float): A number with 3 thousands values after decimal point
        
        Returns:
            (float)  : The rounded number now with 2 decimal places, but using
                        the old round method
        Example:
            >>> _old_round(782.225)
            >>> 782.23
        Additional:
            For more information about rounding at Python 3, go to:
                https://stackoverflow.com/questions/10825926/python-3-x-rounding-behavior

    """
    # * Multiply ou number by 100 since we want to round the 2 decimal place
    new_n = n * 100
    # * Round using the Decimal lib, but using the old method
    n_dec = Decimal(new_n).quantize(Decimal('1'), rounding=ROUND_HALF_UP)
    # * Roll back to original format
    return float(n_dec / 100)

The function works normally when I use only numbers (see function example above), but when I use the Pandas apply or map to perform the calculation in the CMO _ ($ / MWH) column, some values are not rounded correctly ( index 12 ), does anyone know why? Find the DF after calculation bellow:

df['ROUNDED_CMO'] = df['CMO_($/MWH)'].apply(_old_round)
df

Results:

    INIT_DATE  INIT_HOUR  CMO_($/MWH)   NEW_CMO
0   2020-12-01  00:00:00    754.275     754.28
1   2020-12-01  01:00:00    728.130     728.13
2   2020-12-01  02:00:00    722.575     722.58
3   2020-12-01  03:00:00    722.045     722.05
4   2020-12-01  04:00:00    721.950     721.95
5   2020-12-01  05:00:00    721.035     721.04
6   2020-12-01  06:00:00    722.100     722.10
7   2020-12-01  07:00:00    739.925     739.93
8   2020-12-01  08:00:00    771.390     771.39
9   2020-12-01  09:00:00    797.415     797.42
10  2020-12-01  10:00:00    796.585     796.59
11  2020-12-01  11:00:00    791.875     791.88
12  2020-12-01  12:00:00    782.225     782.22 <
13  2020-12-01  13:00:00    783.540     783.54
14  2020-12-01  14:00:00    790.980     790.98
15  2020-12-01  15:00:00    815.555     815.56
16  2020-12-01  16:00:00    824.760     824.76
17  2020-12-01  17:00:00    782.265     782.27
18  2020-12-01  18:00:00    779.970     779.97
19  2020-12-01  19:00:00    784.640     784.64
20  2020-12-01  20:00:00    785.380     785.38
21  2020-12-01  21:00:00    785.840     785.84
22  2020-12-01  22:00:00    779.775     779.78
23  2020-12-01  23:00:00    763.775     763.78

The value at index 12 should be: 782.23 and not 782.22 , the function works for all the data but not for this specific index for some reason... I'm using a Miniconda Environment (Windows 10) with pandas 1.1.3, Python 3.6.12, jupyter-notebook 6.1.4 (I'm running the code here).

Obs: I already tried to use Math.Ceil, Numpy.Ceil, Decimal Package Round Options, and the solutions at: Question 1 , Question 2 , Question 3 , none of them work properly. I also already tried (the index 12 still not rounding):

df['ROUNDED_CMO'] = df['CMO_($/MWH)'].map(lambda x : _old_round(x))
df['ROUNDED_CMO'] = df['CMO_($/MWH)'].apply(lambda x : _old_round(x))

Answer 1

For me your code is working perfectly, maybe you have a version problem in your packages. I will show you:

• My pandas version: '1.0.5'.
• My decimal version: '1.70'

First I create your dataframe:

import pandas as pd
from decimal import Decimal, ROUND_HALF_UP

df = pd.DataFrame({
    "CMO_($/MWH)" :
        [
        754.275,
        728.130,
        722.575,
        722.045,
        721.950,
        721.035,
        722.100,
        739.925,
        771.390,
        797.415,
        796.585,
        791.875,
        782.225,
        783.540,
        790.980,
        815.555,
        824.760,
        782.265,
        779.970,
        784.640,
        785.380,
        785.840,
        779.775,
        763.775
        ]
})

Then I apply your function:

df['ROUNDED_CMO'] = df['CMO_($/MWH)'].map(lambda x : _old_round(x))

The output is:

CMO_($/MWH) ROUNDED_CMO
0   754.275 754.28
1   728.130 728.13
2   722.575 722.58
3   722.045 722.05
4   721.950 721.95
5   721.035 721.04
6   722.100 722.10
7   739.925 739.93
8   771.390 771.39
9   797.415 797.42
10  796.585 796.59
11  791.875 791.88
12  782.225 782.23
13  783.540 783.54
14  790.980 790.98
15  815.555 815.56
16  824.760 824.76
17  782.265 782.27
18  779.970 779.97
19  784.640 784.64
20  785.380 785.38
21  785.840 785.84
22  779.775 779.78
23  763.775 763.78