how to split an iterable in constant-size chunks

Question

I am surprised I could not find a "batch" function that would take as input an iterable and return an iterable of iterables.

For example:

for i in batch(range(0,10), 1): print i
[0]
[1]
...
[9]

or:

for i in batch(range(0,10), 3): print i
[0,1,2]
[3,4,5]
[6,7,8]
[9]

Now, I wrote what I thought was a pretty simple generator:

def batch(iterable, n = 1):
   current_batch = []
   for item in iterable:
       current_batch.append(item)
       if len(current_batch) == n:
           yield current_batch
           current_batch = []
   if current_batch:
       yield current_batch

But the above does not give me what I would have expected:

for x in   batch(range(0,10),3): print x
[0]
[0, 1]
[0, 1, 2]
[3]
[3, 4]
[3, 4, 5]
[6]
[6, 7]
[6, 7, 8]
[9]

So, I have missed something and this probably shows my complete lack of understanding of python generators. Anyone would care to point me in the right direction ?

[Edit: I eventually realized that the above behavior happens only when I run this within ipython rather than python itself]

Answer 1

This is probably more efficient (faster)

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

for x in batch(range(0, 10), 3):
    print x

---

Example using list

data = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] # list of data 

for x in batch(data, 3):
    print(x)

# Output

[0, 1, 2]
[3, 4, 5]
[6, 7, 8]
[9, 10]

It avoids building new lists.

Answer 2

The recipes in the itertools module provide two ways to do this depending on how you want to handle a final odd-sized lot (keep it, pad it with a fillvalue, ignore it, or raise an exception):

from itertools import islice, zip_longest

def batched(iterable, n):
    "Batch data into lists of length n. The last batch may be shorter."
    # batched('ABCDEFG', 3) --> ABC DEF G
    it = iter(iterable)
    while True:
        batch = list(islice(it, n))
        if not batch:
            return
        yield batch

def grouper(iterable, n, *, incomplete='fill', fillvalue=None):
    "Collect data into non-overlapping fixed-length chunks or blocks"
    # grouper('ABCDEFG', 3, fillvalue='x') --> ABC DEF Gxx
    # grouper('ABCDEFG', 3, incomplete='strict') --> ABC DEF ValueError
    # grouper('ABCDEFG', 3, incomplete='ignore') --> ABC DEF
    args = [iter(iterable)] * n
    if incomplete == 'fill':
        return zip_longest(*args, fillvalue=fillvalue)
    if incomplete == 'strict':
        return zip(*args, strict=True)
    if incomplete == 'ignore':
        return zip(*args)
    else:
        raise ValueError('Expected fill, strict, or ignore')

Answer 3

More-itertools includes two functions that do what you need:

• chunked(iterable, n) returns an iterable of lists, each of length n (except the last one, which may be shorter);
• ichunked(iterable, n) is similar, but returns an iterable of iterables instead.

Answer 4

As others have noted, the code you have given does exactly what you want. For another approach using itertools.islice you could see an example of following recipe:

from itertools import islice, chain

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

Answer 5

Solution for Python 3.8 if you are working with iterables that don't define a len function, and get exhausted:

from itertools import islice

def batcher(iterable, batch_size):
    iterator = iter(iterable)
    while batch := list(islice(iterator, batch_size)):
        yield batch

Example usage:

def my_gen():
    yield from range(10)
 
for batch in batcher(my_gen(), 3):
    print(batch)

>>> [0, 1, 2]
>>> [3, 4, 5]
>>> [6, 7, 8]
>>> [9]

Could of course be implemented without the walrus operator as well.

Answer 6

This is a very short code snippet I know that does not use len and works under both Python 2 and 3 (not my creation):

def chunks(iterable, size):
    from itertools import chain, islice
    iterator = iter(iterable)
    for first in iterator:
        yield list(chain([first], islice(iterator, size - 1)))

Answer 7

Weird, seems to work fine for me in Python 2.x

>>> def batch(iterable, n = 1):
...    current_batch = []
...    for item in iterable:
...        current_batch.append(item)
...        if len(current_batch) == n:
...            yield current_batch
...            current_batch = []
...    if current_batch:
...        yield current_batch
...
>>> for x in batch(range(0, 10), 3):
...     print x
...
[0, 1, 2]
[3, 4, 5]
[6, 7, 8]
[9]

Answer 8

A workable version without new features in python 3.8, adapted from @Atra Azami's answer.

import itertools    

def batch_generator(iterable, batch_size=1):
    iterable = iter(iterable)

    while True:
        batch = list(itertools.islice(iterable, batch_size))
        if len(batch) > 0:
            yield batch
        else:
            break

for x in batch_generator(range(0, 10), 3):
    print(x)

Output:

[0, 1, 2]
[3, 4, 5]
[6, 7, 8]
[9]

Answer 9

I like this one,

def batch(x, bs):
    return [x[i:i+bs] for i in range(0, len(x), bs)]

This returns a list of batches of size bs, you can make it a generator by using a generator expression (i for i in iterable) of course.

Answer 10

def batch(iterable, n):
    iterable=iter(iterable)
    while True:
        chunk=[]
        for i in range(n):
            try:
                chunk.append(next(iterable))
            except StopIteration:
                yield chunk
                return
        yield chunk

list(batch(range(10), 3))
[[0, 1, 2], [3, 4, 5], [6, 7, 8], [9]]

Answer 11

Moving as much into CPython as possible, by leveraging islice and iter(callable) behavior:

from itertools import islice

def chunked(generator, size):
    """Read parts of the generator, pause each time after a chunk"""
    # islice returns results until 'size',
    # make_chunk gets repeatedly called by iter(callable).
    gen = iter(generator)
    make_chunk = lambda: list(islice(gen, size))
    return iter(make_chunk, [])

Inspired by more-itertools, and shortened to the essence of that code.

Answer 12

This is what I use in my project. It handles iterables or lists as efficiently as possible.

def chunker(iterable, size):
    if not hasattr(iterable, "__len__"):
        # generators don't have len, so fall back to slower
        # method that works with generators
        for chunk in chunker_gen(iterable, size):
            yield chunk
        return

    it = iter(iterable)
    for i in range(0, len(iterable), size):
        yield [k for k in islice(it, size)]


def chunker_gen(generator, size):
    iterator = iter(generator)
    for first in iterator:

        def chunk():
            yield first
            for more in islice(iterator, size - 1):
                yield more

        yield [k for k in chunk()]

Answer 13

Here is an approach using reduce function.

Oneliner:

from functools import reduce
reduce(lambda cumulator,item: cumulator[-1].append(item) or cumulator if len(cumulator[-1]) < batch_size else cumulator + [[item]], input_array, [[]])

Or more readable version:

from functools import reduce
def batch(input_list, batch_size):
  def reducer(cumulator, item):
    if len(cumulator[-1]) < batch_size:
      cumulator[-1].append(item)
      return cumulator
    else:
      cumulator.append([item])
    return cumulator
  return reduce(reducer, input_list, [[]])

Test:

>>> batch([1,2,3,4,5,6,7], 3)
[[1, 2, 3], [4, 5, 6], [7]]
>>> batch(a, 8)
[[1, 2, 3, 4, 5, 6, 7]]
>>> batch([1,2,3,None,4], 3)
[[1, 2, 3], [None, 4]]

Answer 14

This would work for any iterable.

from itertools import zip_longest, filterfalse

def batch_iterable(iterable, batch_size=2): 
    args = [iter(iterable)] * batch_size 
    return (tuple(filterfalse(lambda x: x is None, group)) for group in zip_longest(fillvalue=None, *args))

It would work like this:

>>>list(batch_iterable(range(0,5)), 2)
[(0, 1), (2, 3), (4,)]

PS: It would not work if iterable has None values.

Answer 15

You can just group iterable items by their batch index.

def batch(items: Iterable, batch_size: int) -> Iterable[Iterable]:
    # enumerate items and group them by batch index
    enumerated_item_groups = itertools.groupby(enumerate(items), lambda t: t[0] // batch_size)
    # extract items from enumeration tuples
    item_batches = ((t[1] for t in enumerated_items) for key, enumerated_items in enumerated_item_groups)
    return item_batches

It is often the case when you want to collect inner iterables so here is more advanced version.

def batch_advanced(items: Iterable, batch_size: int, batches_mapper: Callable[[Iterable], Any] = None) -> Iterable[Iterable]:
    enumerated_item_groups = itertools.groupby(enumerate(items), lambda t: t[0] // batch_size)
    if batches_mapper:
        item_batches = (batches_mapper(t[1] for t in enumerated_items) for key, enumerated_items in enumerated_item_groups)
    else:
        item_batches = ((t[1] for t in enumerated_items) for key, enumerated_items in enumerated_item_groups)
    return item_batches

Examples:

print(list(batch_advanced([1, 9, 3, 5, 2, 4, 2], 4, tuple)))
# [(1, 9, 3, 5), (2, 4, 2)]
print(list(batch_advanced([1, 9, 3, 5, 2, 4, 2], 4, list)))
# [[1, 9, 3, 5], [2, 4, 2]]

Answer 16

Related functionality you may need:

def batch(size, i):
    """ Get the i'th batch of the given size """
    return slice(size* i, size* i + size)

Usage:

>>> [1,2,3,4,5,6,7,8,9,10][batch(3, 1)]
>>> [4, 5, 6]

It gets the i'th batch from the sequence and it can work with other data structures as well, like pandas dataframes (df.iloc[batch(100,0)]) or numpy array (array[batch(100,0)]).

Answer 17

from itertools import *

class SENTINEL: pass

def batch(iterable, n):
    return (tuple(filterfalse(lambda x: x is SENTINEL, group)) for group in zip_longest(fillvalue=SENTINEL, *[iter(iterable)] * n))

print(list(range(10), 3)))
# outputs: [(0, 1, 2), (3, 4, 5), (6, 7, 8), (9,)]
print(list(batch([None]*10, 3)))
# outputs: [(None, None, None), (None, None, None), (None, None, None), (None,)]

Answer 18

I use

def batchify(arr, batch_size):
  num_batches = math.ceil(len(arr) / batch_size)
  return [arr[i*batch_size:(i+1)*batch_size] for i in range(num_batches)]
  

Answer 19

Keep taking (at most) n elements until it runs out.

def chop(n, iterable):
    iterator = iter(iterable)
    while chunk := list(take(n, iterator)):
        yield chunk


def take(n, iterable):
    iterator = iter(iterable)
    for i in range(n):
        try:
            yield next(iterator)
        except StopIteration:
            return

Answer 20

This code has the following features:

• Can take lists or generators (no len()) as input
• Does not require imports of other packages
• No padding added to last batch

def batch_generator(items, batch_size):
    itemid=0 # Keeps track of current position in items generator/list
    batch = [] # Empty batch
    for item in items: 
      batch.append(item) # Append items to batch
      if len(batch)==batch_size:
        yield batch
        itemid += batch_size # Increment the position in items
        batch = []
    yield batch # yield last bit