[](https://github.com/jcmgray/autoray/actions/workflows/tests.yml)
[](https://codecov.io/gh/jcmgray/autoray)
[](https://app.codacy.com/gh/jcmgray/autoray/dashboard?utm_source=gh&utm_medium=referral&utm_content=&utm_campaign=Badge_grade)
[](https://autoray.readthedocs.io)
[](https://pypi.org/project/autoray/)
[](https://anaconda.org/conda-forge/autoray)
[`autoray`](https://autoray.readthedocs.io/en/latest) is a lightweight python AUTOmatic-arRAY library for
abstracting your tensor operations. Primarily it provides an
[*automatic* dispatch mechanism](https://autoray.readthedocs.io/en/latest/automatic_dispatch.html#)
that means you can write backend agnostic code that works for:
* [numpy](https://github.com/numpy/numpy)
* [pytorch](https://pytorch.org/)
* [jax](https://github.com/google/jax)
* [cupy](https://github.com/cupy/cupy)
* [dask](https://github.com/dask/dask)
* [autograd](https://github.com/HIPS/autograd)
* [tensorflow](https://github.com/tensorflow/tensorflow)
* [sparse](https://sparse.pydata.org/)
* [mars](https://github.com/mars-project/mars)
* ... and indeed **any** library that provides a numpy-*ish* api, even if it
knows nothing about `autoray`.
Beyond that, abstracting the array interface allows you to:
* *swap [custom versions of functions](https://autoray.readthedocs.io/en/latest/automatic_dispatch.html#functions)
for specific backends*
* *trace through computations [lazily](https://autoray.readthedocs.io/en/latest/lazy_computation.html) without actually
running them*
* *automatically [share intermediates and fold constants](https://autoray.readthedocs.io/en/latest/lazy_computation.html#sharing-intermediates)
in computations*
* *compile functions with a [unified interface](https://autoray.readthedocs.io/en/latest/compilation.html) for different
backends*
## Basic usage
The main function of `autoray` is
[`do`](https://autoray.readthedocs.io/en/latest/autoapi/autoray/autoray/index.html#autoray.autoray.do),
which takes a function
name followed by `*args` and `**kwargs`, and automatically looks up (and
caches) the correct function to match the equivalent numpy call:
```python
from autoray as ar
def noised_svd(x):
# automatic dispatch based on supplied array
U, s, VH = ar.do('linalg.svd', x)
# automatic dispatch based on different array
sn = s + 0.1 * ar.do('random.normal', size=ar.shape(s), like=s)
# automatic dispatch for multiple arrays for certain functions
return ar.do('einsum', 'ij,j,jk->ik', U, sn, VH)
# explicit backend given by string
x = ar.do('random.uniform', size=(100, 100), like="torch")
# this function now works for any backend
y = noised_svd(x)
# explicit inference of backend from array
ar.infer_backend(y)
# 'torch'
```
If you don't like the explicit `do` syntax, or simply want a
drop-in replacement for existing code, you can also import the `autoray.numpy`
module:
```python
from autoray import numpy as np
# set a temporary default backend
with ar.backend_like('cupy'):
z = np.ones((3, 4), dtype='float32')
np.exp(z)
# array([[2.7182817, 2.7182817, 2.7182817, 2.7182817],
# [2.7182817, 2.7182817, 2.7182817, 2.7182817],
# [2.7182817, 2.7182817, 2.7182817, 2.7182817]], dtype=float32)
```
Custom backends and functions can be dynamically registered with:
* [`register_backend`](https://autoray.readthedocs.io/en/latest/autoapi/autoray/autoray/index.html#autoray.autoray.register_backend)
* [`register_function`](https://autoray.readthedocs.io/en/latest/autoapi/autoray/autoray/index.html#autoray.autoray.register_function)
The main documentation is available at [autoray.readthedocs.io](https://autoray.readthedocs.io/en/latest/).
Raw data
{
"_id": null,
"home_page": null,
"name": "autoray",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.9",
"maintainer_email": null,
"keywords": "agnostic, array, autograd, cupy, dask, dispatch, jax, numeric, numpy, tensor, tensorflow",
"author": null,
"author_email": "Johnnie Gray <johnniemcgray@gmail.com>",
"download_url": "https://files.pythonhosted.org/packages/8e/b7/8ec4ffeca00c9360adb94be177313f711071628b21ea912abe6e246051e1/autoray-0.7.0.tar.gz",
"platform": null,
"description": "\n\n[](https://github.com/jcmgray/autoray/actions/workflows/tests.yml)\n[](https://codecov.io/gh/jcmgray/autoray)\n[](https://app.codacy.com/gh/jcmgray/autoray/dashboard?utm_source=gh&utm_medium=referral&utm_content=&utm_campaign=Badge_grade)\n[](https://autoray.readthedocs.io)\n[](https://pypi.org/project/autoray/)\n[](https://anaconda.org/conda-forge/autoray)\n\n[`autoray`](https://autoray.readthedocs.io/en/latest) is a lightweight python AUTOmatic-arRAY library for\nabstracting your tensor operations. Primarily it provides an\n[*automatic* dispatch mechanism](https://autoray.readthedocs.io/en/latest/automatic_dispatch.html#)\nthat means you can write backend agnostic code that works for:\n\n* [numpy](https://github.com/numpy/numpy)\n* [pytorch](https://pytorch.org/)\n* [jax](https://github.com/google/jax)\n* [cupy](https://github.com/cupy/cupy)\n* [dask](https://github.com/dask/dask)\n* [autograd](https://github.com/HIPS/autograd)\n* [tensorflow](https://github.com/tensorflow/tensorflow)\n* [sparse](https://sparse.pydata.org/)\n* [mars](https://github.com/mars-project/mars)\n* ... and indeed **any** library that provides a numpy-*ish* api, even if it\n knows nothing about `autoray`.\n\nBeyond that, abstracting the array interface allows you to:\n\n* *swap [custom versions of functions](https://autoray.readthedocs.io/en/latest/automatic_dispatch.html#functions)\n for specific backends*\n* *trace through computations [lazily](https://autoray.readthedocs.io/en/latest/lazy_computation.html) without actually\n running them*\n* *automatically [share intermediates and fold constants](https://autoray.readthedocs.io/en/latest/lazy_computation.html#sharing-intermediates)\n in computations*\n* *compile functions with a [unified interface](https://autoray.readthedocs.io/en/latest/compilation.html) for different\n backends*\n\n\n## Basic usage\n\nThe main function of `autoray` is\n[`do`](https://autoray.readthedocs.io/en/latest/autoapi/autoray/autoray/index.html#autoray.autoray.do),\nwhich takes a function\nname followed by `*args` and `**kwargs`, and automatically looks up (and\ncaches) the correct function to match the equivalent numpy call:\n\n```python\nfrom autoray as ar\n\ndef noised_svd(x):\n # automatic dispatch based on supplied array\n U, s, VH = ar.do('linalg.svd', x)\n\n # automatic dispatch based on different array\n sn = s + 0.1 * ar.do('random.normal', size=ar.shape(s), like=s)\n\n # automatic dispatch for multiple arrays for certain functions\n return ar.do('einsum', 'ij,j,jk->ik', U, sn, VH)\n\n# explicit backend given by string\nx = ar.do('random.uniform', size=(100, 100), like=\"torch\")\n\n# this function now works for any backend\ny = noised_svd(x)\n\n# explicit inference of backend from array\nar.infer_backend(y)\n# 'torch'\n```\n\nIf you don't like the explicit `do` syntax, or simply want a\ndrop-in replacement for existing code, you can also import the `autoray.numpy`\nmodule:\n\n```python\nfrom autoray import numpy as np\n\n# set a temporary default backend\nwith ar.backend_like('cupy'):\n z = np.ones((3, 4), dtype='float32')\n\nnp.exp(z)\n# array([[2.7182817, 2.7182817, 2.7182817, 2.7182817],\n# [2.7182817, 2.7182817, 2.7182817, 2.7182817],\n# [2.7182817, 2.7182817, 2.7182817, 2.7182817]], dtype=float32)\n```\n\nCustom backends and functions can be dynamically registered with:\n\n* [`register_backend`](https://autoray.readthedocs.io/en/latest/autoapi/autoray/autoray/index.html#autoray.autoray.register_backend)\n* [`register_function`](https://autoray.readthedocs.io/en/latest/autoapi/autoray/autoray/index.html#autoray.autoray.register_function)\n\nThe main documentation is available at [autoray.readthedocs.io](https://autoray.readthedocs.io/en/latest/).\n",
"bugtrack_url": null,
"license": "Apache",
"summary": "Abstract your array operations.",
"version": "0.7.0",
"project_urls": {
"Changelog": "https://github.com/jcmgray/autoray/releases",
"Documentation": "https://autoray.readthedocs.io/",
"Issues": "https://github.com/jcmgray/autoray/issues",
"Repository": "https://github.com/jcmgray/autoray/"
},
"split_keywords": [
"agnostic",
" array",
" autograd",
" cupy",
" dask",
" dispatch",
" jax",
" numeric",
" numpy",
" tensor",
" tensorflow"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "e5acd8fb343def8bc5b7f82f5dcf0892e9020a446f21107a2d7de1537ff2fdf3",
"md5": "1138b68cc62271fc0dd779d2b1023a21",
"sha256": "03103957df3d1b66b8068158056c2909a72095b19d1b24262261276a714a5d07"
},
"downloads": -1,
"filename": "autoray-0.7.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "1138b68cc62271fc0dd779d2b1023a21",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.9",
"size": 930011,
"upload_time": "2024-10-22T00:13:19",
"upload_time_iso_8601": "2024-10-22T00:13:19.587125Z",
"url": "https://files.pythonhosted.org/packages/e5/ac/d8fb343def8bc5b7f82f5dcf0892e9020a446f21107a2d7de1537ff2fdf3/autoray-0.7.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "8eb78ec4ffeca00c9360adb94be177313f711071628b21ea912abe6e246051e1",
"md5": "a4e2a6f5b8346cb68d984b47af853464",
"sha256": "7829d21258512f87e02f23ce74ae5759af4ce8998069d2cce53468f1d701a219"
},
"downloads": -1,
"filename": "autoray-0.7.0.tar.gz",
"has_sig": false,
"md5_digest": "a4e2a6f5b8346cb68d984b47af853464",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.9",
"size": 1214555,
"upload_time": "2024-10-22T00:13:21",
"upload_time_iso_8601": "2024-10-22T00:13:21.434764Z",
"url": "https://files.pythonhosted.org/packages/8e/b7/8ec4ffeca00c9360adb94be177313f711071628b21ea912abe6e246051e1/autoray-0.7.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-10-22 00:13:21",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "jcmgray",
"github_project": "autoray",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "autoray"
}
JSON
