Python/Pandas：在用户定义函数中使用内置函数作为参数

Question

Very simple query but can't seem to find appropriate answer. I want to pass the Pandas method for eg .sum() as an input to my function.

def something(dataframe,col_name,func):
    return dataframe.col_name.func

something(df,'a',sum())

TypeError: sum expected at least 1 arguments, got 0.

Python confuses it with inbuilt function sum()

Answer 1

You an use operator.methodcaller for this:

from operator import methodcaller

df = pd.DataFrame({'a': range(11)})

def foo(df, col, method):
    return methodcaller(method)(df[col])

res_sum = foo(df, 'a', 'sum')   # 55
res_avg = foo(df, 'a', 'mean')  # 5.0

The reason for your error is you are trying to pass the result of a called function with no arguments, one that happens to require arguments to work.

The benefit of passing strings is you rely on tried-and-tested methods built into the Pandas framework, eg pd.Series.sum , pd.Series.mean , etc. While you can attempt to use Python built-ins and NumPy functions directly with Pandas series, you may find discrepancies versus what you might expect. Stick with documented Pandas methods where possible.

Answer 2

Do not recommend acquiring functions by this method in a general case, but here is a solution without any additional imports. Python has a getattr builtin function which "[r]eturn the value of the named attribute of object." Its usage is getattr(object, name[, default]) . So you need to rewrite your function as the following.

def something(dataframe,col_name,func):
    return getattr(dataframe[col_name], func)

something(df,'a',"sum")

If you want to get the result of the function call sum, replace the function definition with return getattr(dataframe.col_name, func)() .