Python list intersection efficiency: generator or filter()?

Question

I would like to intersect two lists in Python (2.7). I need the result to be iterable:

list1 = [1,2,3,4]
list2 = [3,4,5,6]
result = (3,4) # any kind of iterable

Providing a full iteration will be performed first thing after the intersection, which of the following is more efficient?

Using a generator:

result = (x for x in list1 if x in list2)

Using filter():

result = filter(lambda x: x in list2, list1)

Other suggestions?

Thanks in advance,
Amnon

Answer 1

Neither of these. The best way is to use sets.

list1 = [1,2,3,4]
list2 = [3,4,5,6]
result = set(list1).intersection(list2)

Sets are iterable, so no need to convert the result into anything.

Answer 2

Your solution has a complexity of O(m*n), where m and n are the respective lengths of the two lists. You can improve the complexity to O(m+n) using a set for one of the lists:

s = set(list1)
result = [x for x in list2 if x in s]

In cases where speed matters more than readability (that is, almost never), you can also use

result = filter(set(a).__contains__, b)

which is about 20 percent faster than the other solutions on my machine.

Answer 3

I tried to compare the speed of 3 methods of list intersection:

import random

a = [random.randint(0, 1000) for _ in range(1000)]
b = [random.randint(0, 1000) for _ in range(1000)]

Solution 1: list comprehension

Time elapse: 8.95265507698059

import time
start = time.time()
for _ in range(1000):
    result = [x for x in a if x in b]
elapse = time.time() - start
print(elapse) 

Solution 2: set

Time elapse: 0.09089064598083496

start = time.time()
for _ in range(1000):
    result = set.intersection(set(a), set(b))
elapse = time.time() - start
print(elapse) 

Solution 3: numpy.intersect1d

Time elapse: 0.323300838470459

start = time.time()
for _ in range(1000):
    result = np.intersect1d(a, b)
elapse = time.time() - start
print(elapse) 

Conclusion

I think use set.intersection is the fastest way.