Improving your python code with simple tricks

Sometimes our code is running slow. Sometimes it is eating up memory. Maybe it is just not as readable as we would like it to be. In this post, we will see how to use some functions from the default library to improve our code. All the code used in this post is available here. While I only presented a few functions that I use frequently, there are many more that can be used to improve your code. I encourage you to check the official documentation to see what else is available.

Use list comprehension whenever possible

What does it mean?

List comprehension is basically another way to create a list. Suppose we want to create a list from values in a range:

# traditional way
tmp = []
for i in range(10_000_000):
  tmp.append(i)

# list comprehension
tmp = [i for i in range(10_000_000)]

Why does it matter?

List comprehension is usually much faster than the traditional loop. Let’s compare them:

tmp = []
for i in range(10_000_000):
  tmp.append(i)
1.04 s ± 89.8 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 293.57 MiB
tmp = [i for i in range(10_000_000)]
731 ms ± 71.6 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 285.82 MiB

Other examples

Creating a list with an if condition

tmp = []
for i in range(10_000_000):
  if i % 2 == 0:
    tmp.append(i)
1.11 s ± 24 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: -0.25 MiB
tmp = [i for i in range(10_000_000) if i % 2 == 0]
944 ms ± 6.79 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 115.54 MiB

Creating a list with an if else condition

tmp = []
for i in range(10_000_000):
  if i % 2 == 0:
    tmp.append(i)
  else:
    tmp.append(i+1)
1.61 s ± 90.5 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 32.89 MiB
tmp = [i if i % 2 == 0 else i + 1 for i in range(10_000_000)]
1.33 s ± 11.4 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 273.82 MiB

Fastest way to create a list

When generating a list from a generator (range in this case), it is even faster to use the list() constructor.

tmp = list(range(10_000_000))

To validate this, let’s compare the code for 7 runs, 10 loops each:

  loop list comprehension list constructor
mean ± std. dev. per loop 1.04 s ± 89.8 ms 731 ms ± 71.6 ms 301 ms ± 18.4 ms
memory increment 293.57 MiB 285.82 MiB 75.12 MiB

Why is the list() constructor faster? According to this answer in StackOverflow:

The list comprehension executes the loop in Python bytecode, just like a regular for loop. The list() call iterates entirely in C code, which is far faster.

To compare all these solutions, lets check the equivalent bytecodes. For the loop solution:

1           0 BUILD_LIST               0
            2 STORE_NAME               0 (tmp)

2           4 LOAD_NAME                1 (range)
            6 LOAD_CONST               0 (10000000)
            8 CALL_FUNCTION            1
            10 GET_ITER
    >>   12 FOR_ITER                14 (to 28)
            14 STORE_NAME               2 (i)
            16 LOAD_NAME                0 (tmp)
            18 LOAD_METHOD              3 (append)
            20 LOAD_NAME                2 (i)
            22 CALL_METHOD              1
            24 POP_TOP
            26 JUMP_ABSOLUTE           12
    >>   28 LOAD_CONST               1 (None)
            30 RETURN_VALUE

For the list comprehension solution:

1           0 LOAD_CONST               0 (<code object <listcomp> at 0x7f272c8eaf50, file "<stdin>", line 1>)
            2 LOAD_CONST               1 ('<listcomp>')
            4 MAKE_FUNCTION            0
            6 LOAD_NAME                0 (range)
            8 LOAD_NAME                1 (10_000_000)
            10 CALL_FUNCTION            1
            12 GET_ITER
            14 CALL_FUNCTION            1
            16 POP_TOP
            18 LOAD_CONST               2 (None)
            20 RETURN_VALUE

Disassembly of <code object <listcomp> at 0x7f272c8eaf50, file "<stdin>", line 1>:
1           0 BUILD_LIST               0
            2 LOAD_FAST                0 (.0)
    >>    4 FOR_ITER                 8 (to 14)
            6 STORE_FAST               1 (i)
            8 LOAD_FAST                1 (i)
            10 LIST_APPEND              2
            12 JUMP_ABSOLUTE            4
    >>   14 RETURN_VALUE

And for the list constructor solution:

1           0 LOAD_NAME                0 (list)
            2 LOAD_NAME                1 (range)
            4 LOAD_NAME                2 (10_000_000)
            6 CALL_FUNCTION            1
            8 CALL_FUNCTION            1
            10 POP_TOP
            12 LOAD_CONST               0 (None)
            14 RETURN_VALUE

We can see that the list() constructor generates less bytecodes.

Use generators and iterators whenever possible

To create a generator like a list comprehension (called generator expression), just replace the squared brackets [ ] with parenthesis ( ).

Why does it matter?

Generators looks like a normal function, except that it contains yield expressions for producing a series of values usable in a for-loop or that can be retrieved one at a time with the next() function. It returns a generator iterator, which temporarily suspends processing, remembering the location execution state (including local variables and pending try-statements).

Let’s do some comparisons:

tmp = sum([i for i in range(10_000_000)])
860 ms ± 30.4 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
tmp = sum((i for i in range(10_000_000)))
609 ms ± 2.93 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

Not that different right? Now, let’s check the memory usage. Let’s focus on increment, since it represents the difference in memory between the beginning and end of this execution.

memory increment: 263.44 MiB
memory increment: 0.00 MiB

What happened? The generator only returns one element at a time, which is given to the sum function. This way, we don’t need to pre-generate the whole list to perform the sum of the elements. In fact, since the sum function gets an iterator as parameter, we could call it like this:

tmp = sum(i for i in range(10_000_000))
593 ms ± 90.4 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.01 MiB

Or even:

tmp = sum(range(10_000_000))
168 ms ± 4.68 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.00 MiB

Which runs way faster.

Avoid generating all the values whenever possible

For the sake of these examples, suppose we have an ordered list of values.

What to do if we want only the values lower than a limit?

The list comprehension way of achieving this is this:

tmp = [i for i in range(10_000_000) if i < 1_000_000]
596 ms ± 7.16 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.02 MiB

Now, with loops:

tmp = []
for i in range(10_000_000):
  if i < 1_000_000:
    tmp.append(i)
  else:
    break
116 ms ± 2.37 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.01 MiB

Why is it faster with loops?

Because using list comprehension, the whole list must be generated before selecting the elements. The same is not true for the loop, that only runs through some of the values.

Can we do better?

Yes, with takewhile.

from itertools import takewhile

tmp = list(takewhile(lambda x: x < 1_000_000, range(10_000_000)))
107 ms ± 2.37 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.01 MiB

Note: takewhile is only faster when you know that the condition will be satisfied “soon enough”.


tmp = []
for i in range(10_000_000):
  if i < 9_000_000:
    tmp.append(i)
  else:
    break
1.12 s ± 27.5 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 3.70 MiB
tmp = [i for i in range(10_000_000) if i < 9_000_000]
1.05 s ± 51 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 246.21 MiB
tmp = list(takewhile(lambda x: x < 9_000_000, range(10_000_000)))
1.06 s ± 8.41 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.00 MiB

In this case it is quicker to generate the whole list, and then filter it. But note, while using list comprehension is quicker, takewhile takes less memory, since it still doesn’t need to store the whole list, even momentarily.

What if we want only the values higher than a limit?

First, let’s try with loops:

tmp = []
for i in range(10_000_000):
  if i > 1_000_000:
    tmp.append(i)
1.3 s ± 93.3 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.15 MiB

Now, with list comprehension:

tmp = [i for i in range(10_000_000) if i > 1_000_000]
978 ms ± 11.2 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 169.84 MiB

In this case, since the loop will run through every element, it is slower than the list comprehension. It takes less memory though, since it doesn’t need to store the whole list in memory.

Can we do better?

Yes, with dropwhile.

from itertools import dropwhile

tmp = list(dropwhile(lambda x: x < 1_000_000, range(10_000_000)))
442 ms ± 10.2 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.05 MiB

Note: dropwhile also is only faster when you know that the condition will be satisfied “soon enough”.


tmp = []
for i in range(10_000_000):
  if i > 9_000_000:
    tmp.append(i)
654 ms ± 9.3 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.00 MiB
tmp = [i for i in range(10_000_000) if i > 9_000_000]
623 ms ± 13.6 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.01 MiB
tmp = list(dropwhile(lambda x: x < 9_000_000, range(10_000_000)))
924 ms ± 104 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.01 MiB

What about when we want to get the first N samples?

With loops:

tmp = []
for n, i in enumerate(range(10_000_000)):
  if n < 1_000_000:
    tmp.append(i)
  else:
    break
147 ms ± 3.09 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.00 MiB

Doing it with list comprehension:

tmp = [i for i in range(10_000_000)][:1_000_000]
523 ms ± 10.8 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.01 MiB

Why is it faster with loops?

Because using list comprehension, the whole list must be generated before doing the slice operation. The same is not true for the loop, that only runs through some of the values.

Can we do better?

Yes, with islice.

from itertools import islice

tmp = list(islice((i for i in range(10_000_000)), 1_000_000))
72.5 ms ± 1.82 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.01 MiB

What about when we want to get the last N samples?

We can achieve the same result with islice. Let’s get straight to the comparisons:

tmp = []
for n, i in enumerate(range(10_000_000)):
  if n > 9_000_000:
    tmp.append(i)
1.44 s ± 95.3 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 0.00 MiB
tmp = [i for i in range(10_000_000)][9_000_000:]
743 ms ± 8.52 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: 177.31 MiB
tmp = list(islice((i for i in range(10_000_000)), 9_000_000, None))
796 ms ± 11.8 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
memory increment: -0.16 MiB

Note again that, as it happened with dropwhile when the condition takes longer to be satisfied, while using islice is slower than doing it with list comprehension, it takes much less memory.

What if we just want to count the number of elements that will be generated?

Suppose we want to know how many elements will be generated from a condition. Usually, we would do it like this:

tmp = [i for i in range(value) if i % 2 == 0]
count = len(tmp)
1.02 s ± 63.6 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
increment: 0.04 MiB

But the problem is, we are storing a whole list in memory only to get its length.

Can we do better?

Yes, by creating a generator that generates 1 every time the condition is true, and summing it.

count = sum(1 for i in range(value) if i % 2 == 0)
991 ms ± 18.9 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
increment: 0.00 MiB

Other useful itertools functions

We already introduced 3 of the most useful functions: dropwhile, islice, and takewhile. Let’s check other useful functions.

cycle

Repeats indefinitely a given sequence.

from itertools import cycle

tmp = []

for counter, i in enumerate(cycle(range(4))):
  if counter == 10:
    break

  tmp.append(i)

print(tmp)
[0, 1, 2, 3, 0, 1, 2, 3, 0, 1]

repeat

Repeats indefinitely a given value, unless the times argument is specified.

from itertools import repeat

tmp = list(repeat(10, 5))
print(tmp)
[10, 10, 10, 10, 10]

product

Equivalent to a nested for-loop.

tmp = []

for i in range(2):
  for j in range(2):
    for k in range(2):
      for l in range(2):
        tmp.append(sum([i, j, k, l]))

print(tmp)
[0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4]
from itertools import product

tmp = [sum(i) for i in product(range(2), range(2), range(2), range(2))]
print(tmp)
[0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4]

Since in this case we are using the same sequence for all the loops, we can use repeat to simplify the code:

from itertools import product

tmp = [sum(i) for i in product(range(2), repeat=4)]
print(tmp)
[0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4]

itertools has all the combinatorics functions implemented

product permutation combination w/ replacement combination
AA   AA  
AB AB AB AB
AC AC AC AC
AD AD AD AD
BA BA    
BB   BB  
BC BC BC BC
BD BD BD BD
CA CA    
CB CB    
CC   CC  
CD CD CD CD
DA DA    
DB DB    
DC DC    
DD   DD  

Improving code with functools

Storing function calls with lru_cache

Let’s take the fibonacci function, for example, that calls itself recursively.

def fib(n):
    if n < 2:
        return n
    return fib(n-1) + fib(n-2)

Let’s use it in a list comprehension to get the first 16 fibonacci numbers:

tmp = [fib(i) for i in range(16)]
698 µs ± 152 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

It takes a pretty good amount of time to execute for a small number. But what if we could automatically save the results of the previous calls to the function? That is what lru_cache is all about. It stores previous calls, with its given parameters and calculated output, as a least recent used (LRU) cache. This way, whenever we call the function and this call was already made (and its results are still stored in the cache), we simply get the results from the cache.

from functools import lru_cache

@lru_cache
def fib(n):
    if n < 2:
        return n
    return fib(n-1) + fib(n-2)

Let’s try that line again:

tmp = [fib(i) for i in range(16)]
3.34 µs ± 719 ns per loop (mean ± std. dev. of 7 runs, 10 loops each)

Now, let’s check the cache information:

  • hits: number of times the function was called and the results were already there;
  • misses: number of times the function was called and the results were not there;
  • maxsize: current maximum allowed size of the cache;
  • currsize: actual size of the cache (stored results).
print(fib.cache_info())
CacheInfo(hits=1132, misses=16, maxsize=128, currsize=16)

Creating functions with defaults from partial

Suppose we have a function, called divide_by. It is a pretty generic function, but it is usually called with some specific values, like dividing by two, or by three.

def divide_by(x, y):
  return x / y

print(divide_by(12, 2))
print(divide_by(12, 3))
6.0
4.0

What if, instead of creating an entire new function, we could only create different signatures for the function, one for each common y value? That is what partial is for:

from functools import partial

divide_by_two = partial(divide_by, y=2)
divide_by_three = partial(divide_by, y=3)

print(divide_by_two(12))
print(divide_by_three(12))
6.0
4.0



    Enjoy Reading This Article?

    Here are some more articles you might like to read next:

  • Turning your al-folio into a dual-language website
  • The problem of research code reproducibility
  • Creating localized blog posts
  • sli.dev for non-web developers
  • Creating localized Projects pages