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?
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]]
I like the Python doc's version proposed by tzot and J.F.Sebastian a lot, but it has two shortcomings:
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))
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 [])
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.
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)]
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.
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
As per this answer, the top-voted answer leaves a 'runt' at the end. Here's my solution to really get about as evenly-sized chunks as you can, with no runts. It basically tries to pick exactly the fractional spot where it should split the list, but just rounds it off to the nearest integer:
from __future__ import division # not needed in Python 3
def n_even_chunks(l, n):
"""Yield n as even chunks as possible from l."""
last = 0
for i in range(1, n+1):
cur = int(round(i * (len(l) / n)))
yield l[last:cur]
last = cur
Demonstration:
>>> pprint.pprint(list(n_even_chunks(list(range(100)), 9)))
[[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, 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]]
>>> pprint.pprint(list(n_even_chunks(list(range(100)), 11)))
[[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, 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]]
Compare to the top-voted chunks
answer:
>>> pprint.pprint(list(chunks(list(range(100)), 100//9)))
[[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, 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]]
>>> pprint.pprint(list(chunks(list(range(100)), 100//11)))
[[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, 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]]
def main():
print(chunkify([1,2,3,4,5,6],2))
def chunkify(list, n):
chunks = []
for i in range(0, len(list), n):
chunks.append(list[i:i+n])
return chunks
main()
I think that it's simple and can give you a chunk of an array.
Like @AaronHall I got here looking for roughly evenly sized chunks. There are different interpretations of that. In my case, if the desired size is N, I would like each group to be of size>=N. Thus, the orphans which are created in most of the above should be redistributed to other groups.
This can be done using:
def nChunks(l, n):
""" Yield n successive chunks from l.
Works for lists, pandas dataframes, etc
"""
newn = int(1.0 * len(l) / n + 0.5)
for i in xrange(0, n-1):
yield l[i*newn:i*newn+newn]
yield l[n*newn-newn:]
(from Splitting a list of into N parts of approximately equal length) by simply calling it as nChunks(l,l/n) or nChunks(l,floor(l/n))
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
python pydash
package could be a good choice.
from pydash.arrays import chunk
ids = ['22', '89', '2', '3', '4', '5', '6', '7', '8', '9', '10', '11', '1']
chunk_ids = chunk(ids,5)
print(chunk_ids)
# output: [['22', '89', '2', '3', '4'], ['5', '6', '7', '8', '9'], ['10', '11', '1']]
for more checkout pydash chunk list
[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
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.
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]]
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).
Here's an idea using itertools.groupby:
def chunks(l, n):
c = itertools.count()
return (it for _, it in itertools.groupby(l, lambda x: next(c)//n))
This returns a generator of generators. If you want a list of lists, just replace the last line with
return [list(it) for _, it in itertools.groupby(l, lambda x: next(c)//n)]
Example returning list of lists:
>>> chunks('abcdefghij', 4)
[['a', 'b', 'c', 'd'], ['e', 'f', 'g', 'h'], ['i', 'j']]
(So yes, this suffers form the "runt problem", which may or may not be a problem in a given situation.)
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
.
A generic chunker for any iterable, which gives the user a choice of how to handle a partial chunk at the end.
Tested on Python 3.
chunker.py
from enum import Enum
class PartialChunkOptions(Enum):
INCLUDE = 0
EXCLUDE = 1
PAD = 2
ERROR = 3
class PartialChunkException(Exception):
pass
def chunker(iterable, n, on_partial=PartialChunkOptions.INCLUDE, pad=None):
"""
A chunker yielding n-element lists from an iterable, with various options
about what to do about a partial chunk at the end.
on_partial=PartialChunkOptions.INCLUDE (the default):
include the partial chunk as a short (<n) element list
on_partial=PartialChunkOptions.EXCLUDE
do not include the partial chunk
on_partial=PartialChunkOptions.PAD
pad to an n-element list
(also pass pad=<pad_value>, default None)
on_partial=PartialChunkOptions.ERROR
raise a RuntimeError if a partial chunk is encountered
"""
on_partial = PartialChunkOptions(on_partial)
iterator = iter(iterable)
while True:
vals = []
for i in range(n):
try:
vals.append(next(iterator))
except StopIteration:
if vals:
if on_partial == PartialChunkOptions.INCLUDE:
yield vals
elif on_partial == PartialChunkOptions.EXCLUDE:
pass
elif on_partial == PartialChunkOptions.PAD:
yield vals + [pad] * (n - len(vals))
elif on_partial == PartialChunkOptions.ERROR:
raise PartialChunkException
return
return
yield vals
test.py
import chunker
chunk_size = 3
for it in (range(100, 107),
range(100, 109)):
print("\nITERABLE TO CHUNK: {}".format(it))
print("CHUNK SIZE: {}".format(chunk_size))
for option in chunker.PartialChunkOptions.__members__.values():
print("\noption {} used".format(option))
try:
for chunk in chunker.chunker(it, chunk_size, on_partial=option):
print(chunk)
except chunker.PartialChunkException:
print("PartialChunkException was raised")
print("")
output of test.py
ITERABLE TO CHUNK: range(100, 107)
CHUNK SIZE: 3
option PartialChunkOptions.INCLUDE used
[100, 101, 102]
[103, 104, 105]
[106]
option PartialChunkOptions.EXCLUDE used
[100, 101, 102]
[103, 104, 105]
option PartialChunkOptions.PAD used
[100, 101, 102]
[103, 104, 105]
[106, None, None]
option PartialChunkOptions.ERROR used
[100, 101, 102]
[103, 104, 105]
PartialChunkException was raised
ITERABLE TO CHUNK: range(100, 109)
CHUNK SIZE: 3
option PartialChunkOptions.INCLUDE used
[100, 101, 102]
[103, 104, 105]
[106, 107, 108]
option PartialChunkOptions.EXCLUDE used
[100, 101, 102]
[103, 104, 105]
[106, 107, 108]
option PartialChunkOptions.PAD used
[100, 101, 102]
[103, 104, 105]
[106, 107, 108]
option PartialChunkOptions.ERROR used
[100, 101, 102]
[103, 104, 105]
[106, 107, 108]
In [259]: get_in_chunks = lambda itr,n: ( (v for _,v in g) for _,g in itertools.groupby(enumerate(itr),lambda (ind,_): ind/n)) In [260]: list(list(x) for x in get_in_chunks(range(30),7)) Out[260]: [[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]]
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.
Here is a list of additional approaches:
Given
import itertools as it
import collections as ct
import more_itertools as mit
iterable = range(11)
n = 3
Code
The Standard Library
list(it.zip_longest(*[iter(iterable)] * n))
# [(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, None)]
d = {}
for i, x in enumerate(iterable):
d.setdefault(i//n, []).append(x)
list(d.values())
# [[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10]]
dd = ct.defaultdict(list)
for i, x in enumerate(iterable):
dd[i//n].append(x)
list(dd.values())
# [[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10]]
list(mit.chunked(iterable, n))
# [[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10]]
list(mit.sliced(iterable, n))
# [range(0, 3), range(3, 6), range(6, 9), range(9, 11)]
list(mit.grouper(n, iterable))
# [(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, None)]
list(mit.windowed(iterable, len(iterable)//n, step=n))
# [(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, None)]
References
zip_longest
(related post, related post)setdefault
(ordered results requires Python 3.6+)collections.defaultdict
(ordered results requires Python 3.6+)more_itertools.chunked
(related posted)more_itertools.sliced
more_itertools.grouper
(related post)more_itertools.windowed
(see also stagger
, zip_offset
)+ A third-party library that implements itertools recipes and more. > pip install more_itertools
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]
>>>
Here is a bit of code written in Python3 that does the same as np.array_split.
list(map(list, map(functools.partial(filter, None), itertools.zip_longest(*iter(lambda: tuple(itertools.islice(a, n)), ())))))
It's quite a long one-liner but it does divide the items evenly amongst the resulting sublists.
You could use numpy's array_split function e.g., np.array_split(np.array(data), 20)
to split into 20 nearly equal size chunks.
To make sure chunks are exactly equal in size use np.split
.
No magic, but simple and correct:
def chunks(iterable, n):
"""Yield successive n-sized chunks from iterable."""
values = []
for i, item in enumerate(iterable, 1):
values.append(item)
if i % n == 0:
yield values
values = []
if values:
yield values
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]]
python pydash
package could be a good choice.
from pydash.arrays import chunk
ids = ['22', '89', '2', '3', '4', '5', '6', '7', '8', '9', '10', '11', '1']
chunk_ids = chunk(ids,5)
print(chunk_ids)
# output: [['22', '89', '2', '3', '4'], ['5', '6', '7', '8', '9'], ['10', '11', '1']]
for more checkout pydash chunk list
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.
using List Comprehensions of python
[range(t,t+10) for t in range(1,1000,10)]
[[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],....
....[981, 982, 983, 984, 985, 986, 987, 988, 989, 990],
[991, 992, 993, 994, 995, 996, 997, 998, 999, 1000]]
visit this link to know about List Comprehensions
import pprint pprint.pprint(list(chunks(range(10, 75), 10))) [range(10, 20), range(20, 30), range(30, 40), range(40, 50), range(50, 60), range(60, 70), range(70, 75)]
Confer this implementation's result with the example usage result of the accepted answer.
Many of the above functions assume that the length of the whole iterable are known up front, or at least are cheap to calculate.
For some streamed objects that would mean loading the full data into memory first (e.g. to download the whole file) to get the length information.
If you however don't know the the full size yet, you can use this code instead:
def chunks(iterable, size):
"""
Yield successive chunks from iterable, being `size` long.
https://stackoverflow.com/a/55776536/3423324
:param iterable: The object you want to split into pieces.
:param size: The size each of the resulting pieces should have.
"""
i = 0
while True:
sliced = iterable[i:i + size]
if len(sliced) == 0:
# to suppress stuff like `range(max, max)`.
break
# end if
yield sliced
if len(sliced) < size:
# our slice is not the full length, so we must have passed the end of the iterator
break
# end if
i += size # so we start the next chunk at the right place.
# end while
# end def
This works because the slice command will return less/no elements if you passed the end of an iterable:
"abc"[0:2] == 'ab'
"abc"[2:4] == 'c'
"abc"[4:6] == ''
We now use that result of the slice, and calculate the length of that generated chunk. If it is less than what we expect, we know we can end the iteration.
That way the iterator will not be executed unless access.
With Assignment Expressions in Python 3.8 it becomes quite nice:
import itertools
def batch(iterable, size):
it = iter(iterable)
while item := list(itertools.islice(it, size)):
yield item
This works on an arbitrary iterable, not just a list.
>>> import pprint
>>> pprint.pprint(list(batch(range(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]]
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])]
def chunk(input, size):
return map(None, *([iter(input)] * size))
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]]
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))
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
I realise this question is old (stumbled over it on Google), but surely something like the following is far simpler and clearer than any of the huge complex suggestions and only uses slicing:
def chunker(iterable, chunksize):
for i,c in enumerate(iterable[::chunksize]):
yield iterable[i*chunksize:(i+1)*chunksize]
>>> for chunk in chunker(range(0,100), 10):
... print list(chunk)
...
[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]
... etc ...
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)]
Simple yet elegant
l = range(1, 1000)
print [l[x:x+10] for x in xrange(0, len(l), 10)]
or if you prefer:
def chunks(l, n): return [l[x: x+n] for x in xrange(0, len(l), n)]
chunks(l, 10)
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]
>>>
Here is a bit of code written in Python3 that does the same as np.array_split.
list(map(list, map(functools.partial(filter, None), itertools.zip_longest(*iter(lambda: tuple(itertools.islice(a, n)), ())))))
It's quite a long one-liner but it does divide the items evenly amongst the resulting sublists.
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
.
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
Since I had to do something like this, here's my solution given a generator and a batch size:
def pop_n_elems_from_generator(g, n):
elems = []
try:
for idx in xrange(0, n):
elems.append(g.next())
return elems
except StopIteration:
return elems
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.
"Evenly sized chunks", to me, implies that they are all the same length, or barring that option, at minimal variance in length. E.g. 5 baskets for 21 items could have the following results:
>>> import statistics
>>> statistics.variance([5,5,5,5,1])
3.2
>>> statistics.variance([5,4,4,4,4])
0.19999999999999998
A practical reason to prefer the latter result: 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.
When I originally wrote this answer, none of the other answers were evenly sized chunks - they all leave a runt chunk at the end, so they're not well balanced, and have a higher than necessary variance of lengths.
For example, the current top answer ends with:
[60, 61, 62, 63, 64, 65, 66, 67, 68, 69],
[70, 71, 72, 73, 74]]
Others, like list(grouper(3, range(7)))
, and chunk(range(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?
A high-level balanced solution using itertools.cycle
, which is the way I might do it today. Here's the setup:
from itertools import cycle
items = range(10, 75)
number_of_baskets = 10
Now we need our lists into which to populate the elements:
baskets = [[] for _ in range(number_of_baskets)]
Finally, we zip the elements we're going to allocate together with a cycle of the baskets until we run out of elements, which, semantically, it exactly what we want:
for element, basket in zip(items, cycle(baskets)):
basket.append(element)
Here's the result:
>>> from pprint import pprint
>>> pprint(baskets)
[[10, 20, 30, 40, 50, 60, 70],
[11, 21, 31, 41, 51, 61, 71],
[12, 22, 32, 42, 52, 62, 72],
[13, 23, 33, 43, 53, 63, 73],
[14, 24, 34, 44, 54, 64, 74],
[15, 25, 35, 45, 55, 65],
[16, 26, 36, 46, 56, 66],
[17, 27, 37, 47, 57, 67],
[18, 28, 38, 48, 58, 68],
[19, 29, 39, 49, 59, 69]]
To productionize this solution, we write a function, and provide the type annotations:
from itertools import cycle
from typing import List, Any
def cycle_baskets(items: List[Any], maxbaskets: int) -> List[List[Any]]:
baskets = [[] for _ in range(min(maxbaskets, len(items)))]
for item, basket in zip(items, cycle(baskets)):
basket.append(item)
return baskets
In the above, we take our list of items, and the max number of baskets. We create a list of empty lists, in which to append each element, in a round-robin style.
Another elegant solution is to use slices - specifically the less-commonly used step argument to slices. i.e.:
start = 0
stop = None
step = number_of_baskets
first_basket = items[start:stop:step]
This is especially elegant in that slices don't care how long the data are - the result, our first basket, is only as long as it needs to be. We'll only need to increment the starting point for each basket.
In fact this could be a one-liner, but we'll go multiline for readability and to avoid an overlong line of code:
from typing import List, Any
def slice_baskets(items: List[Any], maxbaskets: int) -> List[List[Any]]:
n_baskets = min(maxbaskets, len(items))
return [items[i::n_baskets] for i in range(n_baskets)]
And islice
from the itertools module will provide a lazily iterating approach, like that which was originally asked for in the question.
I don't expect most use-cases to benefit very much, as the original data is already fully materialized in a list, but for large datasets, it could save nearly half the memory usage.
from itertools import islice
from typing import List, Any, Generator
def yield_islice_baskets(items: List[Any], maxbaskets: int) -> Generator[List[Any], None, None]:
n_baskets = min(maxbaskets, len(items))
for i in range(n_baskets):
yield islice(items, i, None, n_baskets)
View results with:
from pprint import pprint
items = list(range(10, 75))
pprint(cycle_baskets(items, 10))
pprint(slice_baskets(items, 10))
pprint([list(s) for s in yield_islice_baskets(items, 10)])
Here's another balanced solution, adapted from a function I've used in production in the past, that uses the modulo operator:
def baskets_from(items, maxbaskets=25):
baskets = [[] for _ in range(maxbaskets)]
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 range(baskets):
yield [items[y_i] for y_i in range(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 range(baskets):
length = ceiling if x_i < stepdown else floor
yield [items.next() for _ in range(length)]
To test them out:
print(baskets_from(range(6), 8))
print(list(iter_baskets_from(range(6), 8)))
print(list(iter_baskets_contiguous(range(6), 8)))
print(baskets_from(range(22), 8))
print(list(iter_baskets_from(range(22), 8)))
print(list(iter_baskets_contiguous(range(22), 8)))
print(baskets_from('ABCDEFG', 3))
print(list(iter_baskets_from('ABCDEFG', 3)))
print(list(iter_baskets_contiguous('ABCDEFG', 3)))
print(baskets_from(range(26), 5))
print(list(iter_baskets_from(range(26), 5)))
print(list(iter_baskets_contiguous(range(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.
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
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
def chunk(input, size):
return map(None, *([iter(input)] * size))
def chunk(lst):
out = []
for x in xrange(2, len(lst) + 1):
if not len(lst) % x:
factor = len(lst) / x
break
while lst:
out.append([lst.pop(0) for x in xrange(factor)])
return out
You could use numpy's array_split function e.g., np.array_split(np.array(data), 20)
to split into 20 nearly equal size chunks.
To make sure chunks are exactly equal in size use np.split
.
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 [])
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])]
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).
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
The answer above (by koffein) has a little problem: the list is always split into an equal number of splits, not equal number of items per partition. This is my version. The "// chs + 1" takes into account that the number of items may not be divideable exactly by the partition size, so the last partition will only be partially filled.
# Given 'l' is your list
chs = 12 # Your chunksize
partitioned = [ l[i*chs:(i*chs)+chs] for i in range((len(l) // chs)+1) ]
Yes, it is an old question, but I had to post this one, because it is even a little shorter than the similar ones. Yes, the result looks scrambled, but if it is just about even length...
>>> n = 3 # number of groups
>>> biglist = range(30)
>>>
>>> [ biglist[i::n] for i in xrange(n) ]
[[0, 3, 6, 9, 12, 15, 18, 21, 24, 27],
[1, 4, 7, 10, 13, 16, 19, 22, 25, 28],
[2, 5, 8, 11, 14, 17, 20, 23, 26, 29]]
>>> def f(x, n, acc=[]): return f(x[n:], n, acc+[(x[:n])]) if x else acc
>>> f("Hallo Welt", 3)
['Hal', 'lo ', 'Wel', 't']
>>>
If you are into brackets - I picked up a book on Erlang :)
I have come up to following solution without creation temorary list object, which should work with any iterable object. Please note that this version for Python 2.x:
def chunked(iterable, size):
stop = []
it = iter(iterable)
def _next_chunk():
try:
for _ in xrange(size):
yield next(it)
except StopIteration:
stop.append(True)
return
while not stop:
yield _next_chunk()
for it in chunked(xrange(16), 4):
print list(it)
Output:
[0, 1, 2, 3]
[4, 5, 6, 7]
[8, 9, 10, 11]
[12, 13, 14, 15]
[]
As you can see if len(iterable) % size == 0 then we have additional empty iterator object. But I do not think that it is big problem.
I've created these two fancy one-liners which are efficient and lazy, both input and output are iterables, also they doen't depend on any module:
First one-liner is totally lazy meaning that it returns iterator producing iterators (i.e. each chunk produced is iterator iterating over chunk's elements), this version is good for the case if chunks are very large or elements are produced slowly one by one and should become available immediately as they are produced:
chunk_iters = lambda it, n: ((e for i, g in enumerate(((f,), cit)) for j, e in zip(range((1, n - 1)[i]), g)) for cit in (iter(it),) for f in cit)
Second one-liner returns iterator that produces lists. Each list is produced as soon as elements of whole chunk become available through input iterator or if very last element of last chunk is reached. This version should be used if input elements are produced fast or all available immediately. Other wise first more-lazy one-liner version should be used.
chunk_lists = lambda it, n: (l for l in ([],) for i, g in enumerate((it, ((),))) for e in g for l in (l[:len(l) % n] + [e][:1 - i],) if (len(l) % n == 0) != i)
Also I provide multi-line version of first chunk_iters
one-liner, which returns iterator producing another iterators (going through each chunk's elements):
def chunk_iters(it, n):
cit = iter(it)
def one_chunk(f):
yield f
for i, e in zip(range(n - 1), cit):
yield e
for f in cit:
yield one_chunk(f)
With Assignment Expressions in Python 3.8 it becomes quite nice:
import itertools
def batch(iterable, size):
it = iter(iterable)
while item := list(itertools.islice(it, size)):
yield item
This works on an arbitrary iterable, not just a list.
>>> import pprint
>>> pprint.pprint(list(batch(range(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]]
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
>>> def f(x, n, acc=[]): return f(x[n:], n, acc+[(x[:n])]) if x else acc
>>> f("Hallo Welt", 3)
['Hal', 'lo ', 'Wel', 't']
>>>
If you are into brackets - I picked up a book on Erlang :)
def main():
print(chunkify([1,2,3,4,5,6],2))
def chunkify(list, n):
chunks = []
for i in range(0, len(list), n):
chunks.append(list[i:i+n])
return chunks
main()
I think that it's simple and can give you a chunk of an array.
Use list comprehensions:
l = [1,2,3,4,5,6,7,8,9,10,11,12]
k = 5 #chunk size
print [tuple(l[x:y]) for (x, y) in [(x, x+k) for x in range(0, len(l), k)]]
I dislike idea of splitting elements by chunk size, e.g. script can devide 101 to 3 chunks as [50, 50, 1]. For my needs I needed spliting proportionly, and keeping order same. First I wrote my own script, which works fine, and it's very simple. But I've seen later this answer, where script is better than mine, I reccomend it. Here's my script:
def proportional_dividing(N, n):
"""
N - length of array (bigger number)
n - number of chunks (smaller number)
output - arr, containing N numbers, diveded roundly to n chunks
"""
arr = []
if N == 0:
return arr
elif n == 0:
arr.append(N)
return arr
r = N // n
for i in range(n-1):
arr.append(r)
arr.append(N-r*(n-1))
last_n = arr[-1]
# last number always will be r <= last_n < 2*r
# when last_n == r it's ok, but when last_n > r ...
if last_n > r:
# ... and if difference too big (bigger than 1), then
if abs(r-last_n) > 1:
#[2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 7] # N=29, n=12
# we need to give unnecessary numbers to first elements back
diff = last_n - r
for k in range(diff):
arr[k] += 1
arr[-1] = r
# and we receive [3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2]
return arr
def split_items(items, chunks):
arr = proportional_dividing(len(items), chunks)
splitted = []
for chunk_size in arr:
splitted.append(items[:chunk_size])
items = items[chunk_size:]
print(splitted)
return splitted
items = [1,2,3,4,5,6,7,8,9,10,11]
chunks = 3
split_items(items, chunks)
split_items(['a','b','c','d','e','f','g','h','i','g','k','l', 'm'], 3)
split_items(['a','b','c','d','e','f','g','h','i','g','k','l', 'm', 'n'], 3)
split_items(range(100), 4)
split_items(range(99), 4)
split_items(range(101), 4)
and output:
[[1, 2, 3, 4], [5, 6, 7, 8], [9, 10, 11]]
[['a', 'b', 'c', 'd'], ['e', 'f', 'g', 'h'], ['i', 'g', 'k', 'l', 'm']]
[['a', 'b', 'c', 'd', 'e'], ['f', 'g', 'h', 'i', 'g'], ['k', 'l', 'm', 'n']]
[range(0, 25), range(25, 50), range(50, 75), range(75, 100)]
[range(0, 25), range(25, 50), range(50, 75), range(75, 99)]
[range(0, 25), range(25, 50), range(50, 75), range(75, 101)]
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
Here's an idea using itertools.groupby:
def chunks(l, n):
c = itertools.count()
return (it for _, it in itertools.groupby(l, lambda x: next(c)//n))
This returns a generator of generators. If you want a list of lists, just replace the last line with
return [list(it) for _, it in itertools.groupby(l, lambda x: next(c)//n)]
Example returning list of lists:
>>> chunks('abcdefghij', 4)
[['a', 'b', 'c', 'd'], ['e', 'f', 'g', 'h'], ['i', 'j']]
(So yes, this suffers form the "runt problem", which may or may not be a problem in a given situation.)
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
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]]
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.
letting r be the chunk size and L be the initial list, you can do.
chunkL = [ [i for i in L[r*k:r*(k+1)] ] for k in range(len(L)/r)]
Like @AaronHall I got here looking for roughly evenly sized chunks. There are different interpretations of that. In my case, if the desired size is N, I would like each group to be of size>=N. Thus, the orphans which are created in most of the above should be redistributed to other groups.
This can be done using:
def nChunks(l, n):
""" Yield n successive chunks from l.
Works for lists, pandas dataframes, etc
"""
newn = int(1.0 * len(l) / n + 0.5)
for i in xrange(0, n-1):
yield l[i*newn:i*newn+newn]
yield l[n*newn-newn:]
(from Splitting a list of into N parts of approximately equal length) by simply calling it as nChunks(l,l/n) or nChunks(l,floor(l/n))
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]]
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
using List Comprehensions of python
[range(t,t+10) for t in range(1,1000,10)]
[[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],....
....[981, 982, 983, 984, 985, 986, 987, 988, 989, 990],
[991, 992, 993, 994, 995, 996, 997, 998, 999, 1000]]
visit this link to know about List Comprehensions
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]]
Simple yet elegant
l = range(1, 1000)
print [l[x:x+10] for x in xrange(0, len(l), 10)]
or if you prefer:
def chunks(l, n): return [l[x: x+n] for x in xrange(0, len(l), n)]
chunks(l, 10)
I wrote a small library expressly for this purpose, available here. The library's chunked
function is particularly efficient because it's implemented as a generator, so a substantial amount of memory can be saved in certain situations. It also doesn't rely on the slice notation, so any arbitrary iterator can be used.
import iterlib
print list(iterlib.chunked(xrange(1, 1000), 10))
# prints [(1, 2, 3, 4, 5, 6, 7, 8, 9, 10), (11, 12, 13, 14, 15, 16, 17, 18, 19, 20), ...]
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)]
I have one solution below which does work but more important than that solution is a few comments on other approaches. First, a good solution shouldn't require that one loop through the sub-iterators in order. If I run
g = paged_iter(list(range(50)), 11))
i0 = next(g)
i1 = next(g)
list(i1)
list(i0)
The appropriate output for the last command is
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
not
[]
As most of the itertools based solutions here return. This isn't just the usual boring restriction about accessing iterators in order. Imagine a consumer trying to clean up poorly entered data which reversed the appropriate order of blocks of 5, i.e., the data looks like [B5, A5, D5, C5] and should look like [A5, B5, C5, D5] (where A5 is just five elements not a sublist). This consumer would look at the claimed behavior of the grouping function and not hesitate to write a loop like
i = 0
out = []
for it in paged_iter(data,5)
if (i % 2 == 0):
swapped = it
else:
out += list(it)
out += list(swapped)
i = i + 1
This will produce mysteriously wrong results if you sneakily assume that sub-iterators are always fully used in order. It gets even worse if you want to interleave elements from the chunks.
Second, a decent number of the suggested solutions implicitly rely on the fact that iterators have a deterministic order (they don't e.g. set) and while some of the solutions using islice may be ok it worries me.
Third, the itertools grouper approach works but the recipe relies on internal behavior of the zip_longest (or zip) functions that isn't part of their published behavior. In particular, the grouper function only works because in zip_longest(i0...in) the next function is always called in order next(i0), next(i1), ... next(in) before starting over. As grouper passes n copies of the same iterator object it relies on this behavior.
Finally, while the solution below can be improved if you make the assumption criticized above that sub-iterators are accessed in order and fully perused without this assumption one MUST implicitly (via call chain) or explicitly (via deques or other data structure) store elements for each subiterator somewhere. So don't bother wasting time (as I did) assuming one could get around this with some clever trick.
def paged_iter(iterat, n):
itr = iter(iterat)
deq = None
try:
while(True):
deq = collections.deque(maxlen=n)
for q in range(n):
deq.append(next(itr))
yield (i for i in deq)
except StopIteration:
yield (i for i in deq)
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
[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
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).
I like the Python doc's version proposed by tzot and J.F.Sebastian a lot, but it has two shortcomings:
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))
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]]'
If you don't care about the order:
> from itertools import groupby
> batch_no = 3
> data = 'abcdefgh'
> [
[x[1] for x in x[1]]
for x in
groupby(
sorted(
(x[0] % batch_no, x[1])
for x in
enumerate(data)
),
key=lambda x: x[0]
)
]
[['a', 'd', 'g'], ['b', 'e', 'h'], ['c', 'f']]
This solution doesn't generates sets of same size, but distributes values so batches are as big as possible while keeping the number of generated batches.
letting r be the chunk size and L be the initial list, you can do.
chunkL = [ [i for i in L[r*k:r*(k+1)] ] for k in range(len(L)/r)]
This works in v2/v3, is inlineable, generator-based and uses only the standard library:
import itertools
def split_groups(iter_in, group_size):
return ((x for _, x in item) for _, item in itertools.groupby(enumerate(iter_in), key=lambda x: x[0] // group_size))
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]]
import pprint pprint.pprint(list(chunks(range(10, 75), 10))) [range(10, 20), range(20, 30), range(30, 40), range(40, 50), range(50, 60), range(60, 70), range(70, 75)]
Confer this implementation's result with the example usage result of the accepted answer.
Many of the above functions assume that the length of the whole iterable are known up front, or at least are cheap to calculate.
For some streamed objects that would mean loading the full data into memory first (e.g. to download the whole file) to get the length information.
If you however don't know the the full size yet, you can use this code instead:
def chunks(iterable, size):
"""
Yield successive chunks from iterable, being `size` long.
https://stackoverflow.com/a/55776536/3423324
:param iterable: The object you want to split into pieces.
:param size: The size each of the resulting pieces should have.
"""
i = 0
while True:
sliced = iterable[i:i + size]
if len(sliced) == 0:
# to suppress stuff like `range(max, max)`.
break
# end if
yield sliced
if len(sliced) < size:
# our slice is not the full length, so we must have passed the end of the iterator
break
# end if
i += size # so we start the next chunk at the right place.
# end while
# end def
This works because the slice command will return less/no elements if you passed the end of an iterable:
"abc"[0:2] == 'ab'
"abc"[2:4] == 'c'
"abc"[4:6] == ''
We now use that result of the slice, and calculate the length of that generated chunk. If it is less than what we expect, we know we can end the iteration.
That way the iterator will not be executed unless access.
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]]
def chunk(lst):
out = []
for x in xrange(2, len(lst) + 1):
if not len(lst) % x:
factor = len(lst) / x
break
while lst:
out.append([lst.pop(0) for x in xrange(factor)])
return out
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))
I have one solution below which does work but more important than that solution is a few comments on other approaches. First, a good solution shouldn't require that one loop through the sub-iterators in order. If I run
g = paged_iter(list(range(50)), 11))
i0 = next(g)
i1 = next(g)
list(i1)
list(i0)
The appropriate output for the last command is
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
not
[]
As most of the itertools based solutions here return. This isn't just the usual boring restriction about accessing iterators in order. Imagine a consumer trying to clean up poorly entered data which reversed the appropriate order of blocks of 5, i.e., the data looks like [B5, A5, D5, C5] and should look like [A5, B5, C5, D5] (where A5 is just five elements not a sublist). This consumer would look at the claimed behavior of the grouping function and not hesitate to write a loop like
i = 0
out = []
for it in paged_iter(data,5)
if (i % 2 == 0):
swapped = it
else:
out += list(it)
out += list(swapped)
i = i + 1
This will produce mysteriously wrong results if you sneakily assume that sub-iterators are always fully used in order. It gets even worse if you want to interleave elements from the chunks.
Second, a decent number of the suggested solutions implicitly rely on the fact that iterators have a deterministic order (they don't e.g. set) and while some of the solutions using islice may be ok it worries me.
Third, the itertools grouper approach works but the recipe relies on internal behavior of the zip_longest (or zip) functions that isn't part of their published behavior. In particular, the grouper function only works because in zip_longest(i0...in) the next function is always called in order next(i0), next(i1), ... next(in) before starting over. As grouper passes n copies of the same iterator object it relies on this behavior.
Finally, while the solution below can be improved if you make the assumption criticized above that sub-iterators are accessed in order and fully perused without this assumption one MUST implicitly (via call chain) or explicitly (via deques or other data structure) store elements for each subiterator somewhere. So don't bother wasting time (as I did) assuming one could get around this with some clever trick.
def paged_iter(iterat, n):
itr = iter(iterat)
deq = None
try:
while(True):
deq = collections.deque(maxlen=n)
for q in range(n):
deq.append(next(itr))
yield (i for i in deq)
except StopIteration:
yield (i for i in deq)
"Evenly sized chunks", to me, implies that they are all the same length, or barring that option, at minimal variance in length. E.g. 5 baskets for 21 items could have the following results:
>>> import statistics
>>> statistics.variance([5,5,5,5,1])
3.2
>>> statistics.variance([5,4,4,4,4])
0.19999999999999998
A practical reason to prefer the latter result: 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.
When I originally wrote this answer, none of the other answers were evenly sized chunks - they all leave a runt chunk at the end, so they're not well balanced, and have a higher than necessary variance of lengths.
For example, the current top answer ends with:
[60, 61, 62, 63, 64, 65, 66, 67, 68, 69],
[70, 71, 72, 73, 74]]
Others, like list(grouper(3, range(7)))
, and chunk(range(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?
A high-level balanced solution using itertools.cycle
, which is the way I might do it today. Here's the setup:
from itertools import cycle
items = range(10, 75)
number_of_baskets = 10
Now we need our lists into which to populate the elements:
baskets = [[] for _ in range(number_of_baskets)]
Finally, we zip the elements we're going to allocate together with a cycle of the baskets until we run out of elements, which, semantically, it exactly what we want:
for element, basket in zip(items, cycle(baskets)):
basket.append(element)
Here's the result:
>>> from pprint import pprint
>>> pprint(baskets)
[[10, 20, 30, 40, 50, 60, 70],
[11, 21, 31, 41, 51, 61, 71],
[12, 22, 32, 42, 52, 62, 72],
[13, 23, 33, 43, 53, 63, 73],
[14, 24, 34, 44, 54, 64, 74],
[15, 25, 35, 45, 55, 65],
[16, 26, 36, 46, 56, 66],
[17, 27, 37, 47, 57, 67],
[18, 28, 38, 48, 58, 68],
[19, 29, 39, 49, 59, 69]]
To productionize this solution, we write a function, and provide the type annotations:
from itertools import cycle
from typing import List, Any
def cycle_baskets(items: List[Any], maxbaskets: int) -> List[List[Any]]:
baskets = [[] for _ in range(min(maxbaskets, len(items)))]
for item, basket in zip(items, cycle(baskets)):
basket.append(item)
return baskets
In the above, we take our list of items, and the max number of baskets. We create a list of empty lists, in which to append each element, in a round-robin style.
Another elegant solution is to use slices - specifically the less-commonly used step argument to slices. i.e.:
start = 0
stop = None
step = number_of_baskets
first_basket = items[start:stop:step]
This is especially elegant in that slices don't care how long the data are - the result, our first basket, is only as long as it needs to be. We'll only need to increment the starting point for each basket.
In fact this could be a one-liner, but we'll go multiline for readability and to avoid an overlong line of code:
from typing import List, Any
def slice_baskets(items: List[Any], maxbaskets: int) -> List[List[Any]]:
n_baskets = min(maxbaskets, len(items))
return [items[i::n_baskets] for i in range(n_baskets)]
And islice
from the itertools module will provide a lazily iterating approach, like that which was originally asked for in the question.
I don't expect most use-cases to benefit very much, as the original data is already fully materialized in a list, but for large datasets, it could save nearly half the memory usage.
from itertools import islice
from typing import List, Any, Generator
def yield_islice_baskets(items: List[Any], maxbaskets: int) -> Generator[List[Any], None, None]:
n_baskets = min(maxbaskets, len(items))
for i in range(n_baskets):
yield islice(items, i, None, n_baskets)
View results with:
from pprint import pprint
items = list(range(10, 75))
pprint(cycle_baskets(items, 10))
pprint(slice_baskets(items, 10))
pprint([list(s) for s in yield_islice_baskets(items, 10)])
Here's another balanced solution, adapted from a function I've used in production in the past, that uses the modulo operator:
def baskets_from(items, maxbaskets=25):
baskets = [[] for _ in range(maxbaskets)]
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 range(baskets):
yield [items[y_i] for y_i in range(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 range(baskets):
length = ceiling if x_i < stepdown else floor
yield [items.next() for _ in range(length)]
To test them out:
print(baskets_from(range(6), 8))
print(list(iter_baskets_from(range(6), 8)))
print(list(iter_baskets_contiguous(range(6), 8)))
print(baskets_from(range(22), 8))
print(list(iter_baskets_from(range(22), 8)))
print(list(iter_baskets_contiguous(range(22), 8)))
print(baskets_from('ABCDEFG', 3))
print(list(iter_baskets_from('ABCDEFG', 3)))
print(list(iter_baskets_contiguous('ABCDEFG', 3)))
print(baskets_from(range(26), 5))
print(list(iter_baskets_from(range(26), 5)))
print(list(iter_baskets_contiguous(range(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.
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
I don't think I saw this option, so just to add another one :)) :
def chunks(iterable, chunk_size):
i = 0;
while i < len(iterable):
yield iterable[i:i+chunk_size]
i += chunk_size
This question reminds me of the Raku (formerly Perl 6) .comb(n)
method. It breaks up strings into n
-sized chunks. (There's more to it than that, but I'll leave out the details.)
It's easy enough to implement a similar function in Python3 as a lambda expression:
comb = lambda s,n: (s[i:i+n] for i in range(0,len(s),n))
Then you can call it like this:
some_list = list(range(0, 20)) # creates a list of 20 elements
generator = comb(some_list, 4) # creates a generator that will generate lists of 4 elements
for sublist in generator:
print(sublist) # prints a sublist of four elements, as it's generated
Of course, you don't have to assign the generator to a variable; you can just loop over it directly like this:
for sublist in comb(some_list, 4):
print(sublist) # prints a sublist of four elements, as it's generated
As a bonus, this comb()
function also operates on strings:
list( comb('catdogant', 3) ) # returns ['cat', 'dog', 'ant']
Here is a list of additional approaches:
Given
import itertools as it
import collections as ct
import more_itertools as mit
iterable = range(11)
n = 3
Code
The Standard Library
list(it.zip_longest(*[iter(iterable)] * n))
# [(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, None)]
d = {}
for i, x in enumerate(iterable):
d.setdefault(i//n, []).append(x)
list(d.values())
# [[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10]]
dd = ct.defaultdict(list)
for i, x in enumerate(iterable):
dd[i//n].append(x)
list(dd.values())
# [[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10]]
list(mit.chunked(iterable, n))
# [[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10]]
list(mit.sliced(iterable, n))
# [range(0, 3), range(3, 6), range(6, 9), range(9, 11)]
list(mit.grouper(n, iterable))
# [(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, None)]
list(mit.windowed(iterable, len(iterable)//n, step=n))
# [(0, 1, 2), (3, 4, 5), (6, 7, 8), (9, 10, None)]
References
zip_longest
(related post, related post)setdefault
(ordered results requires Python 3.6+)collections.defaultdict
(ordered results requires Python 3.6+)more_itertools.chunked
(related posted)more_itertools.sliced
more_itertools.grouper
(related post)more_itertools.windowed
(see also stagger
, zip_offset
)+ A third-party library that implements itertools recipes and more. > pip install more_itertools
I dislike idea of splitting elements by chunk size, e.g. script can devide 101 to 3 chunks as [50, 50, 1]. For my needs I needed spliting proportionly, and keeping order same. First I wrote my own script, which works fine, and it's very simple. But I've seen later this answer, where script is better than mine, I reccomend it. Here's my script:
def proportional_dividing(N, n):
"""
N - length of array (bigger number)
n - number of chunks (smaller number)
output - arr, containing N numbers, diveded roundly to n chunks
"""
arr = []
if N == 0:
return arr
elif n == 0:
arr.append(N)
return arr
r = N // n
for i in range(n-1):
arr.append(r)
arr.append(N-r*(n-1))
last_n = arr[-1]
# last number always will be r <= last_n < 2*r
# when last_n == r it's ok, but when last_n > r ...
if last_n > r:
# ... and if difference too big (bigger than 1), then
if abs(r-last_n) > 1:
#[2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 7] # N=29, n=12
# we need to give unnecessary numbers to first elements back
diff = last_n - r
for k in range(diff):
arr[k] += 1
arr[-1] = r
# and we receive [3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2]
return arr
def split_items(items, chunks):
arr = proportional_dividing(len(items), chunks)
splitted = []
for chunk_size in arr:
splitted.append(items[:chunk_size])
items = items[chunk_size:]
print(splitted)
return splitted
items = [1,2,3,4,5,6,7,8,9,10,11]
chunks = 3
split_items(items, chunks)
split_items(['a','b','c','d','e','f','g','h','i','g','k','l', 'm'], 3)
split_items(['a','b','c','d','e','f','g','h','i','g','k','l', 'm', 'n'], 3)
split_items(range(100), 4)
split_items(range(99), 4)
split_items(range(101), 4)
and output:
[[1, 2, 3, 4], [5, 6, 7, 8], [9, 10, 11]]
[['a', 'b', 'c', 'd'], ['e', 'f', 'g', 'h'], ['i', 'g', 'k', 'l', 'm']]
[['a', 'b', 'c', 'd', 'e'], ['f', 'g', 'h', 'i', 'g'], ['k', 'l', 'm', 'n']]
[range(0, 25), range(25, 50), range(50, 75), range(75, 100)]
[range(0, 25), range(25, 50), range(50, 75), range(75, 99)]
[range(0, 25), range(25, 50), range(50, 75), range(75, 101)]
Since I had to do something like this, here's my solution given a generator and a batch size:
def pop_n_elems_from_generator(g, n):
elems = []
try:
for idx in xrange(0, n):
elems.append(g.next())
return elems
except StopIteration:
return elems
This works in v2/v3, is inlineable, generator-based and uses only the standard library:
import itertools
def split_groups(iter_in, group_size):
return ((x for _, x in item) for _, item in itertools.groupby(enumerate(iter_in), key=lambda x: x[0] // group_size))
This question reminds me of the Raku (formerly Perl 6) .comb(n)
method. It breaks up strings into n
-sized chunks. (There's more to it than that, but I'll leave out the details.)
It's easy enough to implement a similar function in Python3 as a lambda expression:
comb = lambda s,n: (s[i:i+n] for i in range(0,len(s),n))
Then you can call it like this:
some_list = list(range(0, 20)) # creates a list of 20 elements
generator = comb(some_list, 4) # creates a generator that will generate lists of 4 elements
for sublist in generator:
print(sublist) # prints a sublist of four elements, as it's generated
Of course, you don't have to assign the generator to a variable; you can just loop over it directly like this:
for sublist in comb(some_list, 4):
print(sublist) # prints a sublist of four elements, as it's generated
As a bonus, this comb()
function also operates on strings:
list( comb('catdogant', 3) ) # returns ['cat', 'dog', 'ant']
I realise this question is old (stumbled over it on Google), but surely something like the following is far simpler and clearer than any of the huge complex suggestions and only uses slicing:
def chunker(iterable, chunksize):
for i,c in enumerate(iterable[::chunksize]):
yield iterable[i*chunksize:(i+1)*chunksize]
>>> for chunk in chunker(range(0,100), 10):
... print list(chunk)
...
[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]
... etc ...
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 )]
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]]
If you don't care about the order:
> from itertools import groupby
> batch_no = 3
> data = 'abcdefgh'
> [
[x[1] for x in x[1]]
for x in
groupby(
sorted(
(x[0] % batch_no, x[1])
for x in
enumerate(data)
),
key=lambda x: x[0]
)
]
[['a', 'd', 'g'], ['b', 'e', 'h'], ['c', 'f']]
This solution doesn't generates sets of same size, but distributes values so batches are as big as possible while keeping the number of generated batches.
An abstraction would be
l = [1,2,3,4,5,6,7,8,9]
n = 3
outList = []
for i in range(n, len(l) + n, n):
outList.append(l[i-n:i])
print(outList)
This will print:
[[1, 2, 3], [4, 5, 6], [7, 8, 9]]
Use list comprehensions:
l = [1,2,3,4,5,6,7,8,9,10,11,12]
k = 5 #chunk size
print [tuple(l[x:y]) for (x, y) in [(x, x+k) for x in range(0, len(l), k)]]
An abstraction would be
l = [1,2,3,4,5,6,7,8,9]
n = 3
outList = []
for i in range(n, len(l) + n, n):
outList.append(l[i-n:i])
print(outList)
This will print:
[[1, 2, 3], [4, 5, 6], [7, 8, 9]]
I wrote a small library expressly for this purpose, available here. The library's chunked
function is particularly efficient because it's implemented as a generator, so a substantial amount of memory can be saved in certain situations. It also doesn't rely on the slice notation, so any arbitrary iterator can be used.
import iterlib
print list(iterlib.chunked(xrange(1, 1000), 10))
# prints [(1, 2, 3, 4, 5, 6, 7, 8, 9, 10), (11, 12, 13, 14, 15, 16, 17, 18, 19, 20), ...]
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
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
.
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 )]
def chunk(lst):
out = []
for x in xrange(2, len(lst) + 1):
if not len(lst) % x:
factor = len(lst) / x
break
while lst:
out.append([lst.pop(0) for x in xrange(factor)])
return out
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
.
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
No magic, but simple and correct:
def chunks(iterable, n):
"""Yield successive n-sized chunks from iterable."""
values = []
for i, item in enumerate(iterable, 1):
values.append(item)
if i % n == 0:
yield values
values = []
if values:
yield values
An old school approach that does not require itertools but still works with arbitrary generators:
def chunks(g, n):
"""divide a generator 'g' into small chunks
Yields:
a chunk that has 'n' or less items
"""
n = max(1, n)
buff = []
for item in g:
buff.append(item)
if len(buff) == n:
yield buff
buff = []
if buff:
yield buff
def chunk(lst):
out = []
for x in xrange(2, len(lst) + 1):
if not len(lst) % x:
factor = len(lst) / x
break
while lst:
out.append([lst.pop(0) for x in xrange(factor)])
return out
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
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]]'
As per this answer, the top-voted answer leaves a 'runt' at the end. Here's my solution to really get about as evenly-sized chunks as you can, with no runts. It basically tries to pick exactly the fractional spot where it should split the list, but just rounds it off to the nearest integer:
from __future__ import division # not needed in Python 3
def n_even_chunks(l, n):
"""Yield n as even chunks as possible from l."""
last = 0
for i in range(1, n+1):
cur = int(round(i * (len(l) / n)))
yield l[last:cur]
last = cur
Demonstration:
>>> pprint.pprint(list(n_even_chunks(list(range(100)), 9)))
[[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, 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]]
>>> pprint.pprint(list(n_even_chunks(list(range(100)), 11)))
[[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, 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]]
Compare to the top-voted chunks
answer:
>>> pprint.pprint(list(chunks(list(range(100)), 100//9)))
[[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, 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]]
>>> pprint.pprint(list(chunks(list(range(100)), 100//11)))
[[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, 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]]
Yes, it is an old question, but I had to post this one, because it is even a little shorter than the similar ones. Yes, the result looks scrambled, but if it is just about even length...
>>> n = 3 # number of groups
>>> biglist = range(30)
>>>
>>> [ biglist[i::n] for i in xrange(n) ]
[[0, 3, 6, 9, 12, 15, 18, 21, 24, 27],
[1, 4, 7, 10, 13, 16, 19, 22, 25, 28],
[2, 5, 8, 11, 14, 17, 20, 23, 26, 29]]
I don't think I saw this option, so just to add another one :)) :
def chunks(iterable, chunk_size):
i = 0;
while i < len(iterable):
yield iterable[i:i+chunk_size]
i += chunk_size
I have come up to following solution without creation temorary list object, which should work with any iterable object. Please note that this version for Python 2.x:
def chunked(iterable, size):
stop = []
it = iter(iterable)
def _next_chunk():
try:
for _ in xrange(size):
yield next(it)
except StopIteration:
stop.append(True)
return
while not stop:
yield _next_chunk()
for it in chunked(xrange(16), 4):
print list(it)
Output:
[0, 1, 2, 3]
[4, 5, 6, 7]
[8, 9, 10, 11]
[12, 13, 14, 15]
[]
As you can see if len(iterable) % size == 0 then we have additional empty iterator object. But I do not think that it is big problem.
A generic chunker for any iterable, which gives the user a choice of how to handle a partial chunk at the end.
Tested on Python 3.
chunker.py
from enum import Enum
class PartialChunkOptions(Enum):
INCLUDE = 0
EXCLUDE = 1
PAD = 2
ERROR = 3
class PartialChunkException(Exception):
pass
def chunker(iterable, n, on_partial=PartialChunkOptions.INCLUDE, pad=None):
"""
A chunker yielding n-element lists from an iterable, with various options
about what to do about a partial chunk at the end.
on_partial=PartialChunkOptions.INCLUDE (the default):
include the partial chunk as a short (<n) element list
on_partial=PartialChunkOptions.EXCLUDE
do not include the partial chunk
on_partial=PartialChunkOptions.PAD
pad to an n-element list
(also pass pad=<pad_value>, default None)
on_partial=PartialChunkOptions.ERROR
raise a RuntimeError if a partial chunk is encountered
"""
on_partial = PartialChunkOptions(on_partial)
iterator = iter(iterable)
while True:
vals = []
for i in range(n):
try:
vals.append(next(iterator))
except StopIteration:
if vals:
if on_partial == PartialChunkOptions.INCLUDE:
yield vals
elif on_partial == PartialChunkOptions.EXCLUDE:
pass
elif on_partial == PartialChunkOptions.PAD:
yield vals + [pad] * (n - len(vals))
elif on_partial == PartialChunkOptions.ERROR:
raise PartialChunkException
return
return
yield vals
test.py
import chunker
chunk_size = 3
for it in (range(100, 107),
range(100, 109)):
print("\nITERABLE TO CHUNK: {}".format(it))
print("CHUNK SIZE: {}".format(chunk_size))
for option in chunker.PartialChunkOptions.__members__.values():
print("\noption {} used".format(option))
try:
for chunk in chunker.chunker(it, chunk_size, on_partial=option):
print(chunk)
except chunker.PartialChunkException:
print("PartialChunkException was raised")
print("")
output of test.py
ITERABLE TO CHUNK: range(100, 107)
CHUNK SIZE: 3
option PartialChunkOptions.INCLUDE used
[100, 101, 102]
[103, 104, 105]
[106]
option PartialChunkOptions.EXCLUDE used
[100, 101, 102]
[103, 104, 105]
option PartialChunkOptions.PAD used
[100, 101, 102]
[103, 104, 105]
[106, None, None]
option PartialChunkOptions.ERROR used
[100, 101, 102]
[103, 104, 105]
PartialChunkException was raised
ITERABLE TO CHUNK: range(100, 109)
CHUNK SIZE: 3
option PartialChunkOptions.INCLUDE used
[100, 101, 102]
[103, 104, 105]
[106, 107, 108]
option PartialChunkOptions.EXCLUDE used
[100, 101, 102]
[103, 104, 105]
[106, 107, 108]
option PartialChunkOptions.PAD used
[100, 101, 102]
[103, 104, 105]
[106, 107, 108]
option PartialChunkOptions.ERROR used
[100, 101, 102]
[103, 104, 105]
[106, 107, 108]
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)]
In [259]: get_in_chunks = lambda itr,n: ( (v for _,v in g) for _,g in itertools.groupby(enumerate(itr),lambda (ind,_): ind/n)) In [260]: list(list(x) for x in get_in_chunks(range(30),7)) Out[260]: [[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]]
I've created these two fancy one-liners which are efficient and lazy, both input and output are iterables, also they doen't depend on any module:
First one-liner is totally lazy meaning that it returns iterator producing iterators (i.e. each chunk produced is iterator iterating over chunk's elements), this version is good for the case if chunks are very large or elements are produced slowly one by one and should become available immediately as they are produced:
chunk_iters = lambda it, n: ((e for i, g in enumerate(((f,), cit)) for j, e in zip(range((1, n - 1)[i]), g)) for cit in (iter(it),) for f in cit)
Second one-liner returns iterator that produces lists. Each list is produced as soon as elements of whole chunk become available through input iterator or if very last element of last chunk is reached. This version should be used if input elements are produced fast or all available immediately. Other wise first more-lazy one-liner version should be used.
chunk_lists = lambda it, n: (l for l in ([],) for i, g in enumerate((it, ((),))) for e in g for l in (l[:len(l) % n] + [e][:1 - i],) if (len(l) % n == 0) != i)
Also I provide multi-line version of first chunk_iters
one-liner, which returns iterator producing another iterators (going through each chunk's elements):
def chunk_iters(it, n):
cit = iter(it)
def one_chunk(f):
yield f
for i, e in zip(range(n - 1), cit):
yield e
for f in cit:
yield one_chunk(f)
An old school approach that does not require itertools but still works with arbitrary generators:
def chunks(g, n):
"""divide a generator 'g' into small chunks
Yields:
a chunk that has 'n' or less items
"""
n = max(1, n)
buff = []
for item in g:
buff.append(item)
if len(buff) == n:
yield buff
buff = []
if buff:
yield buff
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).
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.
The answer above (by koffein) has a little problem: the list is always split into an equal number of splits, not equal number of items per partition. This is my version. The "// chs + 1" takes into account that the number of items may not be divideable exactly by the partition size, so the last partition will only be partially filled.
# Given 'l' is your list
chs = 12 # Your chunksize
partitioned = [ l[i*chs:(i*chs)+chs] for i in range((len(l) // chs)+1) ]
Source: Stackoverflow.com