使用Anaconda搭建TensorFlow-GPU环境

　　

前言：

对于深度学习来说，各种框架torch,caffe,keras,mxnet,tensorflow,pandapanda环境要求各一，如果我们在一台服务器上部署了较多的这样的框架，那么各种莫名的冲突

会一直伴随着你，吃过很多次亏之后，慢慢的接触了Anaconda，真的是很爽的一个功能，来管理环境配置。我们进行tensorflow安装的时候，还是使用Anaconda，鉴于国内墙太高

，我们使用了Tsinghua的镜像文件，清华大学的Anaconda介绍地址见：https://mirror.tuna.tsinghua.edu.cn/help/anaconda/

这里记录下linux的安装方式：

　所使用的系统: ubuntu16.10

　　安装步骤
1： 先登录到这个页面：https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/
　　 2. 下载： wget -c https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda2-2.4.1-Linux-x86_64.sh 3. 运行： bash  Anaconda2-2.i.1-Linux-x86_64.sh [中间会有几个询问，全部设置yes或者y]
　 4. 设置镜像仓库：
　　　    　TUNA 还提供了 Anaconda 仓库的镜像，运行以下命令:
　　　　　    　conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/ 　　　　　　    conda config --set show_channel_urls yes
　 　　    即可添加 Anaconda Python 免费仓库。
　　 　    运行 conda install numpy 测试一下吧。
5. 安装tensorflow:
5.1 查询conda下的tensorflow可以利用的镜像：
　　　　　　anaconda search -t conda tensorflow

　　大概会出现这些信息：

gxjun@gxjun:~$ anaconda search -t conda tensorflow
Using Anaconda API: https://api.anaconda.org Run 'anaconda show <USER/PACKAGE>' to get more details:
Packages:
Name                      |  Version | Package Types   | Platforms
------------------------- |   ------ | --------------- | ---------------
HCC/tensorflow            |    1.0.0 | conda           | linux-64
HCC/tensorflow-cpucompat  |    1.0.0 | conda           | linux-64
HCC/tensorflow-fma        |    1.0.0 | conda           | linux-64
SentientPrime/tensorflow  |    0.6.0 | conda           | osx-64
: TensorFlow helps the tensors flow
acellera/tensorflow-cuda  |   0.12.1 | conda           | linux-64
anaconda/tensorflow       |    1.1.0 | conda           | linux-ppc64le, linux-64, osx-64, win-64
anaconda/tensorflow-gpu   |    1.1.0 | conda           | linux-ppc64le, linux-64, win-64
conda-forge/r-tensorflow  |      0.7 | conda           | linux-64, osx-64, win-64
conda-forge/tensorflow    |    1.2.0 | conda           | linux-64, win-64, osx-64
: TensorFlow helps the tensors flow
creditx/tensorflow        |    0.9.0 | conda           | linux-64
: TensorFlow helps the tensors flow
derickl/tensorflow        |    1.1.0 | conda           | osx-64
dhirschfeld/tensorflow    |    1.2.0 | conda           | win-64
: Computation using data flow graphs for scalable machine learning
dseuss/tensorflow         |          | conda           | osx-64
guyanhua/tensorflow       |    1.0.0 | conda           | linux-64
ijstokes/tensorflow       | 2017.03.03.1349 | conda, ipynb    | linux-64
jjh_cio_testing/tensorflow |    1.2.1 | conda           | linux-64
: TensorFlow is a machine learning library
jjh_cio_testing/tensorflow-gpu |    1.2.1 | conda           | linux-64
: TensorFlow is a machine learning library
jjh_ppc64le/tensorflow    |    1.2.1 | conda           | linux-ppc64le
: TensorFlow is a machine learning library
jjh_ppc64le/tensorflow-gpu |    1.2.1 | conda           | linux-ppc64le
: TensorFlow is a machine learning library
jjhelmus/tensorflow       | 0.12.0rc0 | conda, pypi     | linux-64, osx-64
: TensorFlow helps the tensors flow
jjhelmus/tensorflow-gpu   |    1.0.1 | conda           | linux-64
kevin-keraudren/tensorflow |    0.9.0 | conda           | linux-64
lcls-rhel7/tensorflow     |    1.1.0 | conda           | linux-64
marta-sd/tensorflow       |    1.2.0 | conda           | linux-64
marta-sd/tensorflow-gpu   |    1.2.0 | conda           | linux-64
memex/tensorflow          |    0.5.0 | conda           | linux-64, osx-64
: TensorFlow helps the tensors flow
mhworth/tensorflow        |    0.7.1 | conda           | osx-64
: TensorFlow helps the tensors flow
miovision/tensorflow      | 0.10.0.gpu | conda           | linux-64, osx-64
msarahan/tensorflow       | 1.0.0rc2 | conda           | linux-64
mutirri/tensorflow        | 0.10.0rc0 | conda           | linux-64
mwojcikowski/tensorflow   |    1.0.1 | conda           | linux-64
nehaljwani/tensorflow     |    1.1.0 | conda           | win-64, osx-64
: TensorFlow is a machine learning library
nehaljwani/tensorflow-gpu |    1.1.0 | conda           | win-64
: TensorFlow is a machine learning library
rdonnelly/tensorflow      |    0.9.0 | conda           | linux-64
rdonnellyr/r-tensorflow   |    0.4.0 | conda           | osx-64
test_org_002/tensorflow   | 0.10.0rc0 | conda           |
Found 36 packages

我们选择其中的一个进行安装之前，先查询这个分支的URL路径：

gxjun@gxjun:~$ anaconda show  nehaljwani/tensorflow-gpu
Using Anaconda API: https://api.anaconda.org Name:    tensorflow-gpu
Summary: TensorFlow is a machine learning library
Access:  public
Package Types:  conda
Versions:
+ 1.1.0

To install this package with conda run:
conda install --channel https://conda.anaconda.org/nehaljwani tensorflow-gpu

5.2 安装

conda install --channel https://conda.anaconda.org/nehaljwani tensorflow-gpu

5.3 检测是否安装成功：

　　　在控制端输入：　　
　  　　　　　python -> 进入python编辑环境
　　　　　　  import tensorflow as tf

　　如果没有报错，则说明幸运的安装成功了~

　　对于失败的情况，我这里给出最容易出现的：

>>> import tensorflow as tf
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/gxjun/anaconda2/lib/python2.7/site-packages/tensorflow/__init__.py", line 24, in <module>
from tensorflow.python import *
File "/home/gxjun/anaconda2/lib/python2.7/site-packages/tensorflow/python/__init__.py", line 49, in <module>
from tensorflow.python import pywrap_tensorflow
File "/home/gxjun/anaconda2/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow.py", line 52, in <module>
raise ImportError(msg)
ImportError: Traceback (most recent call last):
File "/home/gxjun/anaconda2/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow.py", line 41, in <module>
from tensorflow.python.pywrap_tensorflow_internal import *
File "/home/gxjun/anaconda2/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
_pywrap_tensorflow_internal = swig_import_helper()
File "/home/gxjun/anaconda2/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
_mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
ImportError: libcusolver.so.7.5: cannot open shared object file: No such file or directory

这种问题，是说我们没有找到这个动态库，或者干脆就没有这个动态库.

　　 解决方法：

　　　　　　先问是不是： 输入这条命令查查看有没有： locate libcusolver.so 　　　　　

gxjun@gxjun:~$ locate   libcusolver.so
/usr/lib/x86_64-linux-gnu/libcusolver.so
/usr/lib/x86_64-linux-gnu/libcusolver.so.8.0
/usr/lib/x86_64-linux-gnu/libcusolver.so.8.0.44
/usr/lib/x86_64-linux-gnu/stubs/libcusolver.so
/usr/local/cuda-8.0/doc/man/man7/libcusolver.so.7
/usr/local/cuda-8.0/targets/x86_64-linux/lib/libcusolver.so
/usr/local/cuda-8.0/targets/x86_64-linux/lib/libcusolver.so.8.0
/usr/local/cuda-8.0/targets/x86_64-linux/lib/libcusolver.so.8.0.61
/usr/local/cuda-8.0/targets/x86_64-linux/lib/stubs/libcusolver.so
/usr/share/man/man7/libcusolver.so.7.gz

我们发现我们只有libcusolver.so.8.0，并没有我们要找的libcusolver.so.7.5,看了一下官方的文档：

　　给出的建议是： 可以使用.8.0来替代.7.5，我们命名一个.8.0的软连接为.7.5

我们先到/usr/lib/cuda/lib64 下：

ln -s libcusolver.so.8.0  libcusolver.so.7.5

然后在.bashrc系统环境下配置一下这个路径：

export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/lib/cuda/lib64"
export CUDA_HOME=/usr/local/cuda

然后在测试：

　　　　

gxjun@gxjun:~$ python
Python 2.7.12 |Anaconda 4.2.0 (64-bit)| (default, Jul  2 2016, 17:42:40)
[GCC 4.4.7 20120313 (Red Hat 4.4.7-1)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
Anaconda is brought to you by Continuum Analytics.
Please check out: http://continuum.io/thanks and https://anaconda.org >>> import tensorflow as tf
>>>

正常了，说明已经完全安装好了~

参考：

　　　　https://mirror.tuna.tsinghua.edu.cn/help/anaconda/

　　　　http://www.jianshu.com/p/7be2498785b1
https://stackoverflow.com/questions/42013316/after-building-tensorflow-from-source-seeing-libcudart-so-and-libcudnn-errors https://github.com/tensorflow/tensorflow/issues/1501