Pythonic way to sorting list of namedtuples by field name

I want to sort a list of named tuples without having to remember the index of the fieldname. My solution seems rather awkward and was hoping someone would have a more elegant solution.

from operator import itemgetter
from collections import namedtuple


Person = namedtuple('Person', 'name age score')
seq = [
Person(name='nick', age=23, score=100),
Person(name='bob', age=25, score=200),
]


# sort list by name
print(sorted(seq, key=itemgetter(Person._fields.index('name'))))
# sort list by age
print(sorted(seq, key=itemgetter(Person._fields.index('age'))))

Thanks, Nick

34060 次浏览
from operator import attrgetter
from collections import namedtuple


Person = namedtuple('Person', 'name age score')
seq = [Person(name='nick', age=23, score=100),
Person(name='bob', age=25, score=200)]

Sort list by name

sorted(seq, key=attrgetter('name'))

Sort list by age

sorted(seq, key=attrgetter('age'))
sorted(seq, key=lambda x: x.name)
sorted(seq, key=lambda x: x.age)

I tested the two alternatives given here for speed, since @zenpoy was concerned about performance.

Testing script:

import random
from collections import namedtuple
from timeit import timeit
from operator import attrgetter


runs = 10000
size = 10000
random.seed = 42
Person = namedtuple('Person', 'name,age')
seq = [Person(str(random.randint(0, 10 ** 10)), random.randint(0, 100)) for _ in range(size)]


def attrgetter_test_name():
return sorted(seq.copy(), key=attrgetter('name'))


def attrgetter_test_age():
return sorted(seq.copy(), key=attrgetter('age'))


def lambda_test_name():
return sorted(seq.copy(), key=lambda x: x.name)


def lambda_test_age():
return sorted(seq.copy(), key=lambda x: x.age)


print('attrgetter_test_name', timeit(stmt=attrgetter_test_name, number=runs))
print('attrgetter_test_age', timeit(stmt=attrgetter_test_age, number=runs))
print('lambda_test_name', timeit(stmt=lambda_test_name, number=runs))
print('lambda_test_age', timeit(stmt=lambda_test_age, number=runs))

Results:

attrgetter_test_name 44.26793992166096
attrgetter_test_age 31.98247099677627
lambda_test_name 47.97959511074551
lambda_test_age 35.69356267603864

Using lambda was indeed slower. Up to 10% slower.

EDIT:

Further testing shows the results when sorting using multiple attributes. Added the following two test cases with the same setup:

def attrgetter_test_both():
return sorted(seq.copy(), key=attrgetter('age', 'name'))


def lambda_test_both():
return sorted(seq.copy(), key=lambda x: (x.age, x.name))


print('attrgetter_test_both', timeit(stmt=attrgetter_test_both, number=runs))
print('lambda_test_both', timeit(stmt=lambda_test_both, number=runs))

Results:

attrgetter_test_both 92.80101586919373
lambda_test_both 96.85089983147456

Lambda still underperforms, but less so. Now about 5% slower.

Testing is done on Python 3.6.0.

since nobody mentioned using itemgetter(), here how you do using itemgetter().

from operator import itemgetter
from collections import namedtuple


Person = namedtuple('Person', 'name age score')
seq = [
Person(name='nick', age=23, score=100),
Person(name='bob', age=25, score=200),
]


# sort list by name
print(sorted(seq, key=itemgetter(0)))


# sort list by age
print(sorted(seq, key=itemgetter(1)))

This might be a bit too 'magical' for some, but I'm partial to:

# sort list by name
print(sorted(seq, key=Person.name.fget))

Edit: this assumes namedtuple uses the property() built-in to implement the accessors, because it leverages the fget attribute on such a property (see documentation). This may still be true in some implementations but it seems CPython no longer does that, which I think is related to optimization work referenced in https://bugs.python.org/issue32492 (so, since 3.8). Such fragility is the cost of the "magic" I mentioned; namedtuple certainly doesn't promise to use property().

Writing Person.name.__get__ is better (works before & after the implementation change) but is maybe not worth the arcaneness vs. just writing it more plainly as lambda p: p.name