How do you split a list into evenly sized chunks?

问题:

I have a list of arbitrary length, and I need to split it up into equal size chunks and operate on it. There are some obvious ways to do this, like keeping a counter and two lists, and when the second list fills up, add it to the first list and empty the second list for the next round of data, but this is potentially extremely expensive.

I was wondering if anyone had a good solution to this for lists of any length, e.g. using generators.

I was looking for something useful in itertools but I couldn't find anything obviously useful. Might've missed it, though.

Related question: What is the most “pythonic” way to iterate over a list in chunks?

回答1:

Here's a generator that yields the chunks you want:

def chunks(lst, n):
    """Yield successive n-sized chunks from lst."""
    for i in range(0, len(lst), n):
        yield lst[i:i + n]

---

import pprint
pprint.pprint(list(chunks(range(10, 75), 10)))
[[10, 11, 12, 13, 14, 15, 16, 17, 18, 19],
 [20, 21, 22, 23, 24, 25, 26, 27, 28, 29],
 [30, 31, 32, 33, 34, 35, 36, 37, 38, 39],
 [40, 41, 42, 43, 44, 45, 46, 47, 48, 49],
 [50, 51, 52, 53, 54, 55, 56, 57, 58, 59],
 [60, 61, 62, 63, 64, 65, 66, 67, 68, 69],
 [70, 71, 72, 73, 74]]

---

If you're using Python 2, you should use xrange() instead of range():

def chunks(lst, n):
    """Yield successive n-sized chunks from lst."""
    for i in xrange(0, len(lst), n):
        yield lst[i:i + n]

---

Also you can simply use list comprehension instead of writing a function, though it's a good idea to encapsulate operations like this in named functions so that your code is easier to understand. Python 3:

[lst[i:i + n] for i in range(0, len(lst), n)]

Python 2 version:

[lst[i:i + n] for i in xrange(0, len(lst), n)]

回答2:

If you want something super simple:

def chunks(l, n):
    n = max(1, n)
    return (l[i:i+n] for i in range(0, len(l), n))

Use xrange() instead of range() in the case of Python 2.x

回答3:

Directly from the (old) Python documentation (recipes for itertools):

from itertools import izip, chain, repeat

def grouper(n, iterable, padvalue=None):
    "grouper(3, 'abcdefg', 'x') --> ('a','b','c'), ('d','e','f'), ('g','x','x')"
    return izip(*[chain(iterable, repeat(padvalue, n-1))]*n)

The current version, as suggested by J.F.Sebastian:

#from itertools import izip_longest as zip_longest # for Python 2.x
from itertools import zip_longest # for Python 3.x
#from six.moves import zip_longest # for both (uses the six compat library)

def grouper(n, iterable, padvalue=None):
    "grouper(3, 'abcdefg', 'x') --> ('a','b','c'), ('d','e','f'), ('g','x','x')"
    return zip_longest(*[iter(iterable)]*n, fillvalue=padvalue)

I guess Guido's time machine works—worked—will work—will have worked—was working again.

These solutions work because [iter(iterable)]*n (or the equivalent in the earlier version) creates one iterator, repeated n times in the list. izip_longest then effectively performs a round-robin of "each" iterator; because this is the same iterator, it is advanced by each such call, resulting in each such zip-roundrobin generating one tuple of n items.

回答4:

I know this is kind of old but nobody yet mentioned numpy.array_split:

import numpy as np

lst = range(50)
np.array_split(lst, 5)
# [array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]),
#  array([10, 11, 12, 13, 14, 15, 16, 17, 18, 19]),
#  array([20, 21, 22, 23, 24, 25, 26, 27, 28, 29]),
#  array([30, 31, 32, 33, 34, 35, 36, 37, 38, 39]),
#  array([40, 41, 42, 43, 44, 45, 46, 47, 48, 49])]

回答5:

I'm surprised nobody has thought of using iter's two-argument form:

from itertools import islice

def chunk(it, size):
    it = iter(it)
    return iter(lambda: tuple(islice(it, size)), ())

Demo:

>>> list(chunk(range(14), 3))
[(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, 11), (12, 13)]

This works with any iterable and produces output lazily. It returns tuples rather than iterators, but I think it has a certain elegance nonetheless. It also doesn't pad; if you want padding, a simple variation on the above will suffice:

from itertools import islice, chain, repeat

def chunk_pad(it, size, padval=None):
    it = chain(iter(it), repeat(padval))
    return iter(lambda: tuple(islice(it, size)), (padval,) * size)

Demo:

>>> list(chunk_pad(range(14), 3))
[(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, 11), (12, 13, None)]
>>> list(chunk_pad(range(14), 3, 'a'))
[(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, 11), (12, 13, 'a')]

Like the izip_longest-based solutions, the above always pads. As far as I know, there's no one- or two-line itertools recipe for a function that optionally pads. By combining the above two approaches, this one comes pretty close:

_no_padding = object()

def chunk(it, size, padval=_no_padding):
    if padval == _no_padding:
        it = iter(it)
        sentinel = ()
    else:
        it = chain(iter(it), repeat(padval))
        sentinel = (padval,) * size
    return iter(lambda: tuple(islice(it, size)), sentinel)

Demo:

>>> list(chunk(range(14), 3))
[(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, 11), (12, 13)]
>>> list(chunk(range(14), 3, None))
[(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, 11), (12, 13, None)]
>>> list(chunk(range(14), 3, 'a'))
[(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, 11), (12, 13, 'a')]

I believe this is the shortest chunker proposed that offers optional padding.

As Tomasz Gandor observed, the two padding chunkers will stop unexpectedly if they encounter a long sequence of pad values. Here's a final variation that works around that problem in a reasonable way:

_no_padding = object()
def chunk(it, size, padval=_no_padding):
    it = iter(it)
    chunker = iter(lambda: tuple(islice(it, size)), ())
    if padval == _no_padding:
        yield from chunker
    else:
        for ch in chunker:
            yield ch if len(ch) == size else ch + (padval,) * (size - len(ch))

Demo:

>>> list(chunk([1, 2, (), (), 5], 2))
[(1, 2), ((), ()), (5,)]
>>> list(chunk([1, 2, None, None, 5], 2, None))
[(1, 2), (None, None), (5, None)]

回答6:

Here is a generator that work on arbitrary iterables:

def split_seq(iterable, size):
    it = iter(iterable)
    item = list(itertools.islice(it, size))
    while item:
        yield item
        item = list(itertools.islice(it, size))

Example:

>>> import pprint
>>> pprint.pprint(list(split_seq(xrange(75), 10)))
[[0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
 [10, 11, 12, 13, 14, 15, 16, 17, 18, 19],
 [20, 21, 22, 23, 24, 25, 26, 27, 28, 29],
 [30, 31, 32, 33, 34, 35, 36, 37, 38, 39],
 [40, 41, 42, 43, 44, 45, 46, 47, 48, 49],
 [50, 51, 52, 53, 54, 55, 56, 57, 58, 59],
 [60, 61, 62, 63, 64, 65, 66, 67, 68, 69],
 [70, 71, 72, 73, 74]]

回答7:

def chunk(input, size):
    return map(None, *([iter(input)] * size))

回答8:

Simple yet elegant

l = range(1, 1000)
print [l[x:x+10] for x in xrange(0, len(l), 10)]

or if you prefer:

chunks = lambda l, n: [l[x: x+n] for x in xrange(0, len(l), n)]
chunks(l, 10)

回答9:

I saw the most awesome Python-ish answer in a duplicate of this question:

from itertools import zip_longest

a = range(1, 16)
i = iter(a)
r = list(zip_longest(i, i, i))
>>> print(r)
[(1, 2, 3), (4, 5, 6), (7, 8, 9), (10, 11, 12), (13, 14, 15)]

You can create n-tuple for any n. If a = range(1, 15), then the result will be:

[(1, 2, 3), (4, 5, 6), (7, 8, 9), (10, 11, 12), (13, 14, None)]

If the list is divided evenly, then you can replace zip_longest with zip, otherwise the triplet (13, 14, None) would be lost. Python 3 is used above. For Python 2, use izip_longest.

回答10:

Critique of other answers here:

None of these answers are evenly sized chunks, they all leave a runt chunk at the end, so they're not completely balanced. If you were using these functions to distribute work, you've built-in the prospect of one likely finishing well before the others, so it would sit around doing nothing while the others continued working hard.

For example, the current top answer ends with:

[60, 61, 62, 63, 64, 65, 66, 67, 68, 69],
[70, 71, 72, 73, 74]]

I just hate that runt at the end!

Others, like list(grouper(3, xrange(7))), and chunk(xrange(7), 3) both return: [(0, 1, 2), (3, 4, 5), (6, None, None)]. The None's are just padding, and rather inelegant in my opinion. They are NOT evenly chunking the iterables.

Why can't we divide these better?

My Solution(s)

Here's a balanced solution, adapted from a function I've used in production (Note in Python 3 to replace xrange with range):

def baskets_from(items, maxbaskets=25):
    baskets = [[] for _ in xrange(maxbaskets)] # in Python 3 use range
    for i, item in enumerate(items):
        baskets[i % maxbaskets].append(item)
    return filter(None, baskets) 

And I created a generator that does the same if you put it into a list:

def iter_baskets_from(items, maxbaskets=3):
    '''generates evenly balanced baskets from indexable iterable'''
    item_count = len(items)
    baskets = min(item_count, maxbaskets)
    for x_i in xrange(baskets):
        yield [items[y_i] for y_i in xrange(x_i, item_count, baskets)]

And finally, since I see that all of the above functions return elements in a contiguous order (as they were given):

def iter_baskets_contiguous(items, maxbaskets=3, item_count=None):
    '''
    generates balanced baskets from iterable, contiguous contents
    provide item_count if providing a iterator that doesn't support len()
    '''
    item_count = item_count or len(items)
    baskets = min(item_count, maxbaskets)
    items = iter(items)
    floor = item_count // baskets 
    ceiling = floor + 1
    stepdown = item_count % baskets
    for x_i in xrange(baskets):
        length = ceiling if x_i < stepdown else floor
        yield [items.next() for _ in xrange(length)]

Output

To test them out:

print(baskets_from(xrange(6), 8))
print(list(iter_baskets_from(xrange(6), 8)))
print(list(iter_baskets_contiguous(xrange(6), 8)))
print(baskets_from(xrange(22), 8))
print(list(iter_baskets_from(xrange(22), 8)))
print(list(iter_baskets_contiguous(xrange(22), 8)))
print(baskets_from('ABCDEFG', 3))
print(list(iter_baskets_from('ABCDEFG', 3)))
print(list(iter_baskets_contiguous('ABCDEFG', 3)))
print(baskets_from(xrange(26), 5))
print(list(iter_baskets_from(xrange(26), 5)))
print(list(iter_baskets_contiguous(xrange(26), 5)))

Which prints out:

[[0], [1], [2], [3], [4], [5]]
[[0], [1], [2], [3], [4], [5]]
[[0], [1], [2], [3], [4], [5]]
[[0, 8, 16], [1, 9, 17], [2, 10, 18], [3, 11, 19], [4, 12, 20], [5, 13, 21], [6, 14], [7, 15]]
[[0, 8, 16], [1, 9, 17], [2, 10, 18], [3, 11, 19], [4, 12, 20], [5, 13, 21], [6, 14], [7, 15]]
[[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10, 11], [12, 13, 14], [15, 16, 17], [18, 19], [20, 21]]
[['A', 'D', 'G'], ['B', 'E'], ['C', 'F']]
[['A', 'D', 'G'], ['B', 'E'], ['C', 'F']]
[['A', 'B', 'C'], ['D', 'E'], ['F', 'G']]
[[0, 5, 10, 15, 20, 25], [1, 6, 11, 16, 21], [2, 7, 12, 17, 22], [3, 8, 13, 18, 23], [4, 9, 14, 19, 24]]
[[0, 5, 10, 15, 20, 25], [1, 6, 11, 16, 21], [2, 7, 12, 17, 22], [3, 8, 13, 18, 23], [4, 9, 14, 19, 24]]
[[0, 1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15], [16, 17, 18, 19, 20], [21, 22, 23, 24, 25]]

Notice that the contiguous generator provide chunks in the same length patterns as the other two, but the items are all in order, and they are as evenly divided as one may divide a list of discrete elements.

回答11:

If you know list size:

def SplitList(mylist, chunk_size):
    return [mylist[offs:offs+chunk_size] for offs in range(0, len(mylist), chunk_size)]

If you don't (an iterator):

def IterChunks(sequence, chunk_size):
    res = []
    for item in sequence:
        res.append(item)
        if len(res) >= chunk_size:
            yield res
            res = []
    if res:
        yield res  # yield the last, incomplete, portion

In the latter case, it can be rephrased in a more beautiful way if you can be sure that the sequence always contains a whole number of chunks of given size (i.e. there is no incomplete last chunk).

回答12:

If you had a chunk size of 3 for example, you could do:

zip(*[iterable[i::3] for i in range(3)]) 

source: http://code.activestate.com/recipes/303060-group-a-list-into-sequential-n-tuples/

I would use this when my chunk size is fixed number I can type, e.g. '3', and would never change.

回答13:

The toolz library has the partition function for this:

from toolz.itertoolz.core import partition

list(partition(2, [1, 2, 3, 4]))
[(1, 2), (3, 4)]

回答14:

I like the Python doc's version proposed by tzot and J.F.Sebastian a lot, but it has two shortcomings:

• it is not very explicit
• I usually don't want a fill value in the last chunk

I'm using this one a lot in my code:

from itertools import islice

def chunks(n, iterable):
    iterable = iter(iterable)
    while True:
        yield tuple(islice(iterable, n)) or iterable.next()

UPDATE: A lazy chunks version:

from itertools import chain, islice

def chunks(n, iterable):
   iterable = iter(iterable)
   while True:
       yield chain([next(iterable)], islice(iterable, n-1))

回答15:

At this point, I think we need a recursive generator, just in case...

In python 2:

def chunks(li, n):
    if li == []:
        return
    yield li[:n]
    for e in chunks(li[n:], n):
        yield e

In python 3:

def chunks(li, n):
    if li == []:
        return
    yield li[:n]
    yield from chunks(li[n:], n)

Also, in case of massive Alien invasion, a decorated recursive generator might become handy:

def dec(gen):
    def new_gen(li, n):
        for e in gen(li, n):
            if e == []:
                return
            yield e
    return new_gen

@dec
def chunks(li, n):
    yield li[:n]
    for e in chunks(li[n:], n):
        yield e

回答16:

code:

def split_list(the_list, chunk_size):
    result_list = []
    while the_list:
        result_list.append(the_list[:chunk_size])
        the_list = the_list[chunk_size:]
    return result_list

a_list = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

print split_list(a_list, 3)

result:

[[1, 2, 3], [4, 5, 6], [7, 8, 9], [10]]

回答17:

I was curious about the performance of different approaches and here it is:

Tested on Python 3.5.1

import time
batch_size = 7
arr_len = 298937

#---------slice-------------

print("\r\nslice")
start = time.time()
arr = [i for i in range(0, arr_len)]
while True:
    if not arr:
        break

    tmp = arr[0:batch_size]
    arr = arr[batch_size:-1]
print(time.time() - start)

#-----------index-----------

print("\r\nindex")
arr = [i for i in range(0, arr_len)]
start = time.time()
for i in range(0, round(len(arr) / batch_size + 1)):
    tmp = arr[batch_size * i : batch_size * (i + 1)]
print(time.time() - start)

#----------batches 1------------

def batch(iterable, n=1):
    l = len(iterable)
    for ndx in range(0, l, n):
        yield iterable[ndx:min(ndx + n, l)]

print("\r\nbatches 1")
arr = [i for i in range(0, arr_len)]
start = time.time()
for x in batch(arr, batch_size):
    tmp = x
print(time.time() - start)

#----------batches 2------------

from itertools import islice, chain

def batch(iterable, size):
    sourceiter = iter(iterable)
    while True:
        batchiter = islice(sourceiter, size)
        yield chain([next(batchiter)], batchiter)


print("\r\nbatches 2")
arr = [i for i in range(0, arr_len)]
start = time.time()
for x in batch(arr, batch_size):
    tmp = x
print(time.time() - start)

#---------chunks-------------
def chunks(l, n):
    """Yield successive n-sized chunks from l."""
    for i in range(0, len(l), n):
        yield l[i:i + n]
print("\r\nchunks")
arr = [i for i in range(0, arr_len)]
start = time.time()
for x in chunks(arr, batch_size):
    tmp = x
print(time.time() - start)

#-----------grouper-----------

from itertools import zip_longest # for Python 3.x
#from six.moves import zip_longest # for both (uses the six compat library)

def grouper(iterable, n, padvalue=None):
    "grouper(3, 'abcdefg', 'x') --> ('a','b','c'), ('d','e','f'), ('g','x','x')"
    return zip_longest(*[iter(iterable)]*n, fillvalue=padvalue)

arr = [i for i in range(0, arr_len)]
print("\r\ngrouper")
start = time.time()
for x in grouper(arr, batch_size):
    tmp = x
print(time.time() - start)

Results:

slice
31.18285083770752

index
0.02184295654296875

batches 1
0.03503894805908203

batches 2
0.22681021690368652

chunks
0.019841909408569336

grouper
0.006506919860839844

回答18:

[AA[i:i+SS] for i in range(len(AA))[::SS]]

Where AA is array, SS is chunk size. For example:

>>> AA=range(10,21);SS=3
>>> [AA[i:i+SS] for i in range(len(AA))[::SS]]
[[10, 11, 12], [13, 14, 15], [16, 17, 18], [19, 20]]
# or [range(10, 13), range(13, 16), range(16, 19), range(19, 21)] in py3

回答19:

You may also use get_chunks function of utilspie library as:

>>> from utilspie import iterutils
>>> a = [1, 2, 3, 4, 5, 6, 7, 8, 9]

>>> list(iterutils.get_chunks(a, 5))
[[1, 2, 3, 4, 5], [6, 7, 8, 9]]

You can install utilspie via pip:

sudo pip install utilspie

Disclaimer: I am the creator of utilspie library.

回答20:

heh, one line version

In [48]: chunk = lambda ulist, step:  map(lambda i: ulist[i:i+step],  xrange(0, len(ulist), step))

In [49]: chunk(range(1,100), 10)
Out[49]: 
[[1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
 [11, 12, 13, 14, 15, 16, 17, 18, 19, 20],
 [21, 22, 23, 24, 25, 26, 27, 28, 29, 30],
 [31, 32, 33, 34, 35, 36, 37, 38, 39, 40],
 [41, 42, 43, 44, 45, 46, 47, 48, 49, 50],
 [51, 52, 53, 54, 55, 56, 57, 58, 59, 60],
 [61, 62, 63, 64, 65, 66, 67, 68, 69, 70],
 [71, 72, 73, 74, 75, 76, 77, 78, 79, 80],
 [81, 82, 83, 84, 85, 86, 87, 88, 89, 90],
 [91, 92, 93, 94, 95, 96, 97, 98, 99]]

回答21:

def split_seq(seq, num_pieces):
    start = 0
    for i in xrange(num_pieces):
        stop = start + len(seq[i::num_pieces])
        yield seq[start:stop]
        start = stop

usage:

seq = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

for seq in split_seq(seq, 3):
    print seq

回答22:

Another more explicit version.

def chunkList(initialList, chunkSize):
    """
    This function chunks a list into sub lists 
    that have a length equals to chunkSize.

    Example:
    lst = [3, 4, 9, 7, 1, 1, 2, 3]
    print(chunkList(lst, 3)) 
    returns
    [[3, 4, 9], [7, 1, 1], [2, 3]]
    """
    finalList = []
    for i in range(0, len(initialList), chunkSize):
        finalList.append(initialList[i:i+chunkSize])
    return finalList

回答23:

Without calling len() which is good for large lists:

def splitter(l, n):
    i = 0
    chunk = l[:n]
    while chunk:
        yield chunk
        i += n
        chunk = l[i:i+n]

And this is for iterables:

def isplitter(l, n):
    l = iter(l)
    chunk = list(islice(l, n))
    while chunk:
        yield chunk
        chunk = list(islice(l, n))

The functional flavour of the above:

def isplitter2(l, n):
    return takewhile(bool,
                     (tuple(islice(start, n))
                            for start in repeat(iter(l))))

OR:

def chunks_gen_sentinel(n, seq):
    continuous_slices = imap(islice, repeat(iter(seq)), repeat(0), repeat(n))
    return iter(imap(tuple, continuous_slices).next,())

OR:

def chunks_gen_filter(n, seq):
    continuous_slices = imap(islice, repeat(iter(seq)), repeat(0), repeat(n))
    return takewhile(bool,imap(tuple, continuous_slices))

回答24:

One more solution

def make_chunks(data, chunk_size): 
    while data:
        chunk, data = data[:chunk_size], data[chunk_size:]
        yield chunk

>>> for chunk in make_chunks([1, 2, 3, 4, 5, 6, 7], 2):
...     print chunk
... 
[1, 2]
[3, 4]
[5, 6]
[7]
>>> 

回答25:

def chunks(iterable,n):
    """assumes n is an integer>0
    """
    iterable=iter(iterable)
    while True:
        result=[]
        for i in range(n):
            try:
                a=next(iterable)
            except StopIteration:
                break
            else:
                result.append(a)
        if result:
            yield result
        else:
            break

g1=(i*i for i in range(10))
g2=chunks(g1,3)
print g2
'<generator object chunks at 0x0337B9B8>'
print list(g2)
'[[0, 1, 4], [9, 16, 25], [36, 49, 64], [81]]'

回答26:

Consider using matplotlib.cbook pieces

for example:

import matplotlib.cbook as cbook
segments = cbook.pieces(np.arange(20), 3)
for s in segments:
     print s

回答27:

See this reference

>>> orange = range(1, 1001)
>>> otuples = list( zip(*[iter(orange)]*10))
>>> print(otuples)
[(1, 2, 3, 4, 5, 6, 7, 8, 9, 10), ... (991, 992, 993, 994, 995, 996, 997, 998, 999, 1000)]
>>> olist = [list(i) for i in otuples]
>>> print(olist)
[[1, 2, 3, 4, 5, 6, 7, 8, 9, 10], ..., [991, 992, 993, 994, 995, 996, 997, 998, 999, 1000]]
>>> 

Python3

回答28:

a = [1, 2, 3, 4, 5, 6, 7, 8, 9]
CHUNK = 4
[a[i*CHUNK:(i+1)*CHUNK] for i in xrange((len(a) + CHUNK - 1) / CHUNK )]

回答29:

At this point, I think we need the obligatory anonymous-recursive function.

Y = lambda f: (lambda x: x(x))(lambda y: f(lambda *args: y(y)(*args)))
chunks = Y(lambda f: lambda n: [n[0][:n[1]]] + f((n[0][n[1]:], n[1])) if len(n[0]) > 0 else [])

回答30:

Since everybody here talking about iterators. boltons has perfect method for that, called iterutils.chunked_iter.

from boltons import iterutils

list(iterutils.chunked_iter(list(range(50)), 11))

Output:

[[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
 [11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21],
 [22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32],
 [33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43],
 [44, 45, 46, 47, 48, 49]]

But if you don't want to be mercy on memory, you can use old-way and store the full list in the first place with iterutils.chunked.