Window上安装Tensorflow

Anaconda

Anaconda是由Python提供支持的领先的开放数据科学平台。 Anaconda的开源版本是Python和R的高性能分发版本，包括超过100个用于数据科学的最流行的Python，R和Scala软件包。

操作说明：

1. Managing conda

conda --versionconda update conda

2. Managing environments

Create and activate an environment

conda create --name snowflakes biopython

This will create a new environment named /envs/snowflakes with the program Biopython.

Activate the new environment:

• Linux,OS X: source activate snowflakes

• Windows: activate snowflakes

Create a second environment

conda create --name bunnies python=3 astroid babel

This will create a second new environment named /envs/bunnies with Python 3 and Astroid and Babel installed.

List all environments

conda info --envs

Switch to another environment (activate/deactivate)

• Linux,OS X: source activate bunnies

• Windows: activate bunnies

3. Managing Python

Check Python versions

conda search --full-name python

Install a different version of Python

conda create --name snakes python=3

• Linux,OS X: source activate snowflakes

• Windows: activate snowflakes

4. Managing packages

View a list of packages and versions installed in an environment

conda list

Search for a package

conda search beautifulsoup4

Install a new package

conda install --name bunnies beautifulsoup4

Install a package from Anaconda.org

conda install --channel https://conda.anaconda.org/pandas bottleneck

Install a package with pip

pip install see

安装tensorflow

目前tensorflow对python3.5支持友好

1. 创建名为tensorflow的环境

conda create --name tensorflow python=3.5

2. 激活环境

activate tensorflow

3. 安装tensorflow gpu版本包

-- CPU版本pip install tensorflow-- GPU版本pip install tensorflow-gpu  

安装CUDA

不安装CUDA Driver会出现如下错误：

ImportError: No module named '_pywrap_tensorflow_internal'

安装CUDA Tools

安装cuDNN

注意：需要下载5.1版本，6.0版本不可用

Cudnn解压后将bin,include,lib三个文件夹里面的内容覆盖至Cuda安装目录下，默认路径为C:\Program Files\NVIDIA GPUComputing Toolkit\CUDA\v8.0（记住不是替换，是把Cudnn文件里的.dll文件添加到Cuda里面）

接下来，检验Cuda是否安装成功：
进入cuda路径，默认C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\extras\demo_suite执行里面的deviceQuery.exe文件

No module named ‘matplotlib’

conda install matplotlib

Anaconda上查看

第一个tensorflow程序

import tensorflow as tfimport numpy as npimport matplotlib.pyplot as plt%matplotlib inlinea = tf.random_normal([2, 20])sess = tf.Session()out = sess.run(a)x, y = outplt.scatter(x,y)plt.show()