What arguments does Python sort() function have?

Arguments of sort and sorted

Both sort and sorted have three keyword arguments: cmp, key and reverse.

L.sort(cmp=None, key=None, reverse=False) -- stable sort *IN PLACE*;
cmp(x, y) -> -1, 0, 1

sorted(iterable, cmp=None, key=None, reverse=False) --> new sorted list

Using key and reverse is preferred, because they work much faster than an equivalent cmp.

key should be a function which takes an item and returns a value to compare and sort by. reverse allows to reverse sort order.

Using key argument

You can use operator.itemgetter as a key argument to sort by second, third etc. item in a tuple.

Example

>>> from operator import itemgetter

>>> a = range(5)
>>> b = a[::-1]
>>> c = map(lambda x: chr(((x+3)%5)+97), a)
>>> sequence = zip(a,b,c)

# sort by first item in a tuple
>>> sorted(sequence, key = itemgetter(0))
[(0, 4, 'd'), (1, 3, 'e'), (2, 2, 'a'), (3, 1, 'b'), (4, 0, 'c')]

# sort by second item in a tuple
>>> sorted(sequence, key = itemgetter(1))
[(4, 0, 'c'), (3, 1, 'b'), (2, 2, 'a'), (1, 3, 'e'), (0, 4, 'd')]

# sort by third item in a tuple
>>> sorted(sequence, key = itemgetter(2))
[(2, 2, 'a'), (3, 1, 'b'), (4, 0, 'c'), (0, 4, 'd'), (1, 3, 'e')]

Explanation

Sequences can contain any objects, not even comparable, but if we can define a function which produces something we can compare for each of the items, we can pass this function in key argument to sort or sorted.

itemgetter, in particular, creates such a function that fetches the given item from its operand. An example from its documentation:

After, f=itemgetter(2), the call f(r) returns r[2].

Mini-benchmark, key vs cmp

Just out of curiosity, key and cmp performance compared, smaller is better:

>>> from timeit import Timer
>>> Timer(stmt="sorted(xs,key=itemgetter(1))",setup="from operator import itemgetter;xs=range(100);xs=zip(xs,xs);").timeit(300000)
6.7079150676727295
>>> Timer(stmt="sorted(xs,key=lambda x:x[1])",setup="xs=range(100);xs=zip(xs,xs);").timeit(300000)
11.609490871429443
>>> Timer(stmt="sorted(xs,cmp=lambda a,b: cmp(a[1],b[1]))",setup="xs=range(100);xs=zip(xs,xs);").timeit(300000)
22.335839986801147

So, sorting with key seems to be at least twice as fast as sorting with cmp. Using itemgetter instead of lambda x: x[1] makes sort even faster.