2025年12月26日 (五) 17:57 imported>ExultantEditor

2025-12-26T17:57:15Z

新页面

{{Infobox software
| name = Theano
| logo = File:Theano_logo.svg
| logo_size = 260px
| author = [[蒙特利尔大学]]的{{le|蒙特利尔学习算法研究所|Montreal Institute for Learning Algorithms}}（MILA）
| developer = [[PyMC]]开发团队
| released = {{Start date and age|df=yes|2007}}
| latest release version =
| latest release date =
| repo =
| programming language = [[Python]], [[CUDA]]
| platform = [[Linux]], [[macOS]], [[Windows]]
| genre = [[机器学习]], [[函式库]]
| license = [[BSD许可证|3条款BSD许可证]]
| website =
}}

'''Theano'''及其分叉'''PyTensor'''，是一个[[Python]]库和优化的编译器，用来操纵和求值数学表达式特别是矩阵值表达式<ref>{{cite journal |last=Bergstra |first=J. |author2=O. Breuleux |author3=F. Bastien |author4=P. Lamblin |author5=R. Pascanu |author6=G. Desjardins |author7=J. Turian |author8=D. Warde-Farley |author9=Y. Bengio |title=Theano: A CPU and GPU Math Expression Compiler |journal=Proceedings of the Python for Scientific Computing Conference (SciPy) 2010 |date=30 June 2010 |url=http://www.iro.umontreal.ca/~lisa/pointeurs/theano_scipy2010.pdf |access-date=2020-11-06 |archive-date=2020-11-01 |archive-url=https://web.archive.org/web/20201101042841/http://www.iro.umontreal.ca/~lisa/pointeurs/theano_scipy2010.pdf |dead-url=no }}</ref>。在其中，计算使用[[NumPy]]风格的语法来表达并被[[编译器|编译]]，用来在CPU或者[[图形处理器通用计算|GPU]]架构上高效的运行。

==历史==
Theano是[[开放源代码|开源]]项目<ref>{{cite web|title=Github Repository|url=https://github.com/Theano/Theano/|accessdate=2020-11-06|archive-date=2020-11-16|archive-url=https://web.archive.org/web/20201116203703/https://github.com/Theano/Theano|dead-url=no}}</ref>，主要由[[蒙特利尔大学]]的{{le|蒙特利尔学习算法研究所|Montreal Institute for Learning Algorithms}}（MILA）开发<ref>{{cite web|url=http://deeplearning.net/|title=deeplearning.net|accessdate=2020-11-06|archive-date=2017-12-13|archive-url=https://web.archive.org/web/20171213184650/http://deeplearning.net/|dead-url=yes}}</ref>。软件名字取自古代哲学家{{le|Theano (哲学家)|Theano (philosopher)|Theano}}。在2017年9月28日，Pascal Lamblin发布了来自[[约书亚·本希奥]]的一则信息，MILA负责人说：由于更强大的工业参与者的竞争，主要的开发在1.0发行之后将会停止<ref>{{cite mailing list |url=https://groups.google.com/forum/#!topic/theano-users/7Poq8BZutbY |title=MILA and the future of Theano |date=28 September 2017 |accessdate=28 September 2017 |mailing-list=theano-users |last=Lamblin |first=Pascal |archive-date=2011-01-22 |archive-url=http://arquivo.pt/wayback/20110122130054/https://groups.google.com/forum/#!topic/theano-users/7Poq8BZutbY |dead-url=no }}</ref>。Theano 1.0.0随后在2017年11月15日发行<ref>{{cite web|url=http://deeplearning.net/software/theano/NEWS.html|title=Release Notes – Theano 1.0.0 documentation|accessdate=2020-11-06|archive-date=2020-09-14|archive-url=https://web.archive.org/web/20200914204342/http://deeplearning.net/software/theano/NEWS.html|dead-url=yes}}</ref>。

在2018年5月17日，Chris Fonnesbeck代表[[PyMC]]开发团队写道：PyMC开发者将在他们退场后取得对Theano维护的控制权<ref>{{Cite web|url=https://medium.com/@pymc_devs/theano-tensorflow-and-the-future-of-pymc-6c9987bb19d5|title=Theano, TensorFlow and the Future of PyMC|last=Developers|first=PyMC|date=2019-06-01|website=Medium|language=en|access-date=2019-08-27|archive-date=2020-08-06|archive-url=https://web.archive.org/web/20200806150843/https://medium.com/@pymc_devs/theano-tensorflow-and-the-future-of-pymc-6c9987bb19d5|dead-url=no}}</ref>。在2021年1月绝大部份的Theano代码基被重新建造，并增加了通过[[JAX]]和[[Numba]]的编译，修订后的这个计算后端以新名字Aesara发行。2022年11月28日，PyMC团队宣布采用从Aesara计划分叉出PyTensor<ref>{{cite web|url=https://www.pymc.io/blog/pytensor_announcement.html#pytensor_announcement|title=PyMC forked Aesara to PyTensor|access-date=2023-08-17|archive-date=2023-07-18|archive-url=https://web.archive.org/web/20230718090637/https://www.pymc.io/blog/pytensor_announcement.html#pytensor_announcement|dead-url=no}}</ref>。

==样例代码==
下列代码以PyTensor用作介绍的例子：
<syntaxhighlight lang="python">
import pytensor
from pytensor import tensor as pt

# 声明2个符号浮点标量
a = pt.dscalar("a")
b = pt.dscalar("b")

# 建立一个简单的表达式
c = a + b

# 将这个表达式转换成一个可调用对象，
# 它接收'(a, b)'值作为输入并计算出一个值给'c'
f_c = pytensor.function([a, b], c)

assert f_c(1.5, 2.5) == 4.0

# 计算样例表达式关于'a'的梯度
dc = pytensor.grad(c, a)

f_dc = pytensor.function([a, b], dc)

assert f_dc(1.5, 2.5) == 1.0
</syntaxhighlight>
<syntaxhighlight lang="pycon">
>>> import pytensor
>>> from pytensor import tensor as pt
>>>
>>> # 通过'pytensor.function'编译函数还能优化表达式图
>>> # 它会移除不必要的运算并将特定运算替代为更有效的运算
>>>
>>> v = pt.vector("v")
>>> M = pt.matrix("M")
>>>
>>> d = a/a + (M + a).dot(v)
>>>
>>> pytensor.dprint(d)
Add [id A]
├─ ExpandDims{axis=0} [id B]
│ └─ True_div [id C]
│ ├─ a [id D]
│ └─ a [id D]
└─ dot [id E]
├─ Add [id F]
│ ├─ M [id G]
│ └─ ExpandDims{axes=[0, 1]} [id H]
│ └─ a [id D]
└─ v [id I]
<_io.TextIOWrapper name='<stdout>' mode='w' encoding='utf-8'>
>>>
>>> f_d = pytensor.function([a, v, M], d)
>>>
>>> # 'a/a' -> '1'而点积被替代为BLAS函数(i.e. CGemv)
>>> pytensor.dprint(f_d)
Add [id A] 5
├─ [1.] [id B]
└─ CGemv{inplace} [id C] 4
├─ AllocEmpty{dtype='float64'} [id D] 3
│ └─ Shape_i{0} [id E] 2
│ └─ M [id F]
├─ 1.0 [id G]
├─ Add [id H] 1
│ ├─ M [id F]
│ └─ ExpandDims{axes=[0, 1]} [id I] 0
│ └─ a [id J]
├─ v [id K]
└─ 0.0 [id L]
<_io.TextIOWrapper name='<stdout>' mode='w' encoding='utf-8'>
</syntaxhighlight>

==参见==
* {{le|深度学习软件比较|Comparison of deep learning software}}
* [[可微分编程]]

==引用==
{{Reflist|2}}

==外部链接==
* {{Official website|https://github.com/aesara-devs/aesara}} (GitHub)
* [http://deeplearning.net/software/theano/ Theano]{{Wayback|url=http://deeplearning.net/software/theano/ |date=20201108233358 }} at Deep Learning, Université de Montréal

{{深度学习软件}}
{{Differentiable computing|state=collapsed}}

[[Category:阵列编程语言]]
[[Category:深度学习]]
[[Category:自由科学软件]]
[[Category:数值分析语言]]
[[Category:Python科学库]]
[[Category:使用BSD许可证的软件]]

Theano - 版本历史

2025年12月26日 (五) 17:57 imported>ExultantEditor