简体繁体 English

Keras安装的Tensorflow会覆盖Tensorflow-GPU安装

[英]Tensorflow installed by Keras overrides Tensorflow-GPU Installation

原文 2018-10-10 06:16:51 3 1 python/ tensorflow/ keras/ anaconda

I am having a slight issue while trying to make use of keras.preprocessing.text.Tokenizer class. 我在尝试使用keras.preprocessing.text.Tokenizer类时遇到一个小问题。 I've recently installed Tensorflow-GPU and everything worked perfectly fine until I tried to install Keras. 我最近安装了Tensorflow-GPU，在我尝试安装Keras之前，一切工作都很好。

When I installed Keras using Anaconda, I saw that it installed Tensorflow 1.3 as well. 当我使用Anaconda安装Keras时，我看到它也安装了Tensorflow 1.3。 Thing is, it did not DOWNGRADE previously installed Tensorflow-GPU version. 事实是，它没有降级以前安装的Tensorflow-GPU版本。 Now that I try to run my python code, nothing happens, here is screenshot (Note: there should be some printing and processing happening): 现在，我尝试运行python代码，没有任何反应，这是屏幕截图（注意：应该进行一些打印和处理）：

When I try to run python shell and check for Tensorflow version, I get the one that Keras installed. 当我尝试运行python shell并检查Tensorflow版本时，得到了Keras安装的版本。 So I tried to install Tensorflow-GPU again, but it saids that all requirements have been satisfied. 因此，我尝试再次安装Tensorflow-GPU，但它表示已满足所有要求。

How do I get the functionality I need from Keras, but still keep my Tensorflow-GPU installed? 如何从Keras获得所需的功能，但仍然保持Tensorflow-GPU的安装状态？ And why is my code not even trying to run even on Tensorflow 1.3 that Keras installed with it? 为什么我的代码甚至不尝试在Keras所安装的Tensorflow 1.3上运行？

1 个解决方案

I have fixed the issue by using pip install keras instead of conda install -c anaconda keras . 我通过使用pip install keras而不是conda install -c anaconda keras解决了该问题。 By using pip I am only installing keras, and not dependencies that Anaconda includes with it. 通过使用pip，我仅安装keras，而不安装Anaconda附带的依赖项。