📜  TensorFlow与Theano

📅  最后修改于: 2021-01-11 10:59:57             🧑  作者: Mango

TensorFlow和Theano之间的区别

深度学习领域的研究人员使用TensorFlow和Theano,并且经常将它们的受欢迎程度,技术优势,易用性等进行比较。

TensorFlow是由Google Brain团队的工程师和研究人员开发的,而Theano是由蒙特利尔大学MILA的Yoshua Bengio开发的,自2007年以来一直存在。

Theano和TensorFlow之间的主要区别:

两者都是服务于几乎相同目的的库。以下是一些重要差异,如下所述:

  • Theano由隶属于蒙特利尔大学的LISA集团开发,而Google Brain团队开发了TensorFlow供内部使用。通过设计用于内部使用,此后已公开。
  • 当应用程序需要的资源不足并且计算并不复杂时,theano是首选。在开发需要适度系统配置的算法时,毫无疑问可以使用Theano。当需要进行大量计算且资源适当可用时,首选TensorFlow。 TensorFlow的一个优点是,它允许复杂的算法在系统中运行。
  • Theano库提供了仅基于Python的应用程序可以对其进行授权的地方。由于这些限制,对使用C++感兴趣的研究人员不喜欢它。 TensorFlow允许我们将其与C++和Python一起使用,最终为研究提供了扩展的环境。
  • 两者都是为相同的目标而开发的,但是由于组织的作用,它们具有可靠性的标签。由Google开发的Google有一个名为大脑小组的专门团队,并将不断进行开发。
  • TensorFlow比Theano受欢迎得多。
  • Theano由一个名为LISA的团队开发,效果非常好,但是由于其某些局限性,它不如TensorFlow受欢迎。
  • 在流行情况下,TensorFlow赢得了高分,这是事实,它得到了领先的技术传奇之一Google的支持。相比之下,到目前为止,Theano由蒙特利尔大学的MILA积极开发和维护,团队中有50多名成员,为不断改进它做出了贡献。
  • 许多专家认为TensorFlow就像Theano的完全重新实现一样,并且Google创建了TensorFlow来替代Theano。即使如此,后者在许多方面也被认为比TensorFlow更快。
  • Theano还支持更广泛的操作和各种程序,尽管TF在这些方面已显示出希望,但仍需要时间与Theano相提并论。

Theano和TensorFlow的比较

Theano TensorFlow
It is a python based library
Theano is a fully python based library, which means it has to be used with the only python. This library will work with the python language and depends on python programming to be implemented.
It is C++ and python based library
TensorFlow is a C++ and python based library that means it can be used in both the C++ and python programming. Being able to deliver in two languages, it is considered more by the developers.
It uses Single CPU
It uses a single CPU for processing or performing the computations. It makes efficient use of a single CPU and generates the outcome, which is based on the processing power of the CPU.
Uses one or more CPUs
TensorFlow is capable of using one or more CPUs based on how it has been performing. Using a multiple CPU over single has a preference as it leads to reduce the time it may take to complete computations.
Moderate compile speed
Theano is keen to perform complex computations, but sometimes it is not able to meet the requirement due to its low compilation speed. The compile-time is too high but leads to taking time if the program complexity of the program is more elevated.
Fast compile speed
TensorFlow is considered for taking less compilation time as compared to Theano. The fact is that it could make use of multiple CPUs makes it the one that can do complex computations in some time than what is taken by Theano.
Moderate popularity
As compared to TensorFlow, it is considered less popular due to limitations in its feature. It can be used in the python programming and limited to use a single CPU hence only preferred where regular computations are required.
Higher popular
The TensorFlow library has been developed with python and C++. It is capable of working with many CPUs. Due to these popular and preferred at the place that needs complex computations.