Home¶
Overview¶
batcharray
is lightweight library built on top of NumPy
to manipulate nested data structure with NumPy arrays.
This library provides functions for arrays where the first axis is the batch axis.
It also provides functions for arrays representing a batch of sequences where the first axis
is the batch axis and the second axis is the sequence axis.
Motivation¶
Let's imagine you have a batch which is represented by a dictionary with three arrays, and you want
to take the first 2 items.
batcharray
provides the function slice_along_batch
that allows to slide all the arrays:
>>> import numpy as np
>>> from batcharray.nested import slice_along_batch
>>> batch = {
... "a": np.array([[2, 6], [0, 3], [4, 9], [8, 1], [5, 7]]),
... "b": np.array([4, 3, 2, 1, 0]),
... "c": np.array([1.0, 2.0, 3.0, 4.0, 5.0]),
... }
>>> slice_along_batch(batch, stop=2)
{'a': array([[2, 6], [0, 3]]), 'b': array([4, 3]), 'c': array([1., 2.])}
Similarly, it is possible to split a batch in multiple batches by using the
function split_along_batch
:
>>> import numpy as np
>>> from batcharray.nested import split_along_batch
>>> batch = {
... "a": np.array([[2, 6], [0, 3], [4, 9], [8, 1], [5, 7]]),
... "b": np.array([4, 3, 2, 1, 0]),
... "c": np.array([1.0, 2.0, 3.0, 4.0, 5.0]),
... }
>>> split_along_batch(batch, split_size_or_sections=2)
[{'a': array([[2, 6], [0, 3]]), 'b': array([4, 3]), 'c': array([1., 2.])},
{'a': array([[4, 9], [8, 1]]), 'b': array([2, 1]), 'c': array([3., 4.])},
{'a': array([[5, 7]]), 'b': array([0]), 'c': array([5.])}]
Please check the documentation to see all the implemented functions.
API stability¶
While
batcharray
is in development stage, no API is guaranteed to be stable from one
release to the next.
In fact, it is very likely that the API will change multiple times before a stable 1.0.0 release.
In practice, this means that upgrading batcharray
to a new version will possibly break any code
that was using the old version of batcharray
.
License¶
batcharray
is licensed under BSD 3-Clause "New" or "Revised" license available
in LICENSE file.