熊猫groupby（）和agg（）忽略错误

Question

UPDATED for completeness:

import pandas as pd

dates = pd.to_datetime(['2017-10-01','2017-10-02','2017-10-03']).tolist()

df = pd.DataFrame({ 
            'day_of_week':['m','t','w'],
            'alpha':[1,2,3],
            'bravo'[4,5,6],
            'charlie':[7,8,9],
            'dates':dates
            })

agg_dik = {'alpha': sum,
           'bravo': sum,
           'charlie': max,
           'dates': sum}

df = df.groupby('day_of_week').agg(agg_dik).reset_index(drop = True)

And this throws an error on the sum of the datetimes. So I can avoid that if the dataframe truly has five columns, but I have dataframes with hundreds of columns and often build aggregate dictionary comprehensions like:

agg_dik = { c : max if 'e' in c else sum for c in cols }

However, when the groupby().agg() hits a series where sum is not allowed, it errors out.

So my question - is there a way to achieve the results I'm looking for but have pandas either drop the erroring columns or replace with NaN and continue on?

I've looked at a few other questions (like this one ), but they don't fully answer my question.

Answer 1

There are two issues at hand:

1. Your dictionary of functions may contain columns that are not in the dataframe you're working with. In cases like that you will need to grab only the elements whose keys match the columns present in the dataframe.

2. Some of your functions throw errors/exceptions that need to be caught. Otherwise, that list line of your code will not work.

The following is a solution that should handle these two cases:

import pandas as pd
import numpy as np

dates = pd.to_datetime(['2017-10-01','2017-10-02','2017-10-03'])

df = pd.DataFrame({ 
            'day_of_week': ['m','t','w'],
            'alpha': [1,2,3],
            'bravo': [4,5,6],
            'charlie': [7,8,9],
            'dates':dates
            })

def sum_(x):
    try:
        return np.sum(x)
    except:
        return np.nan

def max_(x):
    try:
        return np.max(x)
    except:
        return np.nan

agg_dik = {'alpha': sum_,
           'bravo': sum_,
           'charlie': max_,
           'delta': max_}

df = df.groupby('day_of_week').agg({k:v for k,v in agg_dik.items() if k in df}).reset_index(drop = True)

I hope this helps.