CentOS Anaconda(python3.6)安裝tensorflow

原本安裝tensorflow是一件無比簡單的事，但在個人電腦上卻裝了一個星期。期間遇到各類麻煩事、各類坑，在此記錄一下，方便你們。報錯包括：

• undefined symbol: zgelsd_
• ImportError: cannot import name ‘multiarray’
• whl is not a supported wheel

1，安裝Anaconda

下載地址：https://www.continuum.io/downloads/（我安裝的是linux-64-python3.6) 
我一開始是直接在python上裝，可是python3.4(和python3.5）的numpy版本（1.12.0）彷佛有問題，tensorflow能夠安裝成功，可是運行時調用numpy便報錯了。報錯以下：

import numpy Traceback (most recent call last): File "", line 1, in File "/usr/local/lib/python3.4/dist-packages/numpy/init.py", line 142, in from . import add_newdocs File "/usr/local/lib/python3.4/dist-packages/numpy/add_newdocs.py", line 13, in from numpy.lib import add_newdoc File "/usr/local/lib/python3.4/dist-packages/numpy/lib/init.py", line 18, in from .polynomial import * File "/usr/local/lib/python3.4/dist-packages/numpy/lib/polynomial.py", line 20, in from numpy.linalg import eigvals, lstsq, inv File "/usr/local/lib/python3.4/dist-packages/numpy/linalg/init.py", line 51, in from .linalg import * File "/usr/local/lib/python3.4/dist-packages/numpy/linalg/linalg.py", line 29, in from numpy.linalg import lapack_lite, _umath_linalg ImportError: /usr/local/lib/python3.4/dist-packages/numpy/linalg/lapack_lite.cpython-34m.so: undefined symbol: zgelsd_

在github https://github.com/numpy/numpy/issues/8697上也提問了，可是也沒有解決個人問題 ：undefined symbol: zgelsd_。（期間還出現過ImportError: cannot import name ‘multiarray’ 這種問題，對於linux菜鳥徹底不知道怎麼辦） 
這是numpy的問題，與tensorflow無關，可是我也遲遲沒法解決。無果，轉向直接安裝anaconda，裝好以後，numpy能夠正常運行，tensorflow的安裝卻無比曲折。

2，安裝tensorflow(cpu版）

對anaconda命令的熟悉，能夠參考http://www.jianshu.com/p/d2e15200ee9b 
官方的建議是即時你有gpu，但也能夠先裝一個cpu版，建立環境的命令爲：

conda create -n tensorflow python=3.6 
（必定要指定python版本，我一開始沒有寫python=3.6，後面各類失敗）

先下載安裝包，下載路徑爲：https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.0.0-cp36-cp36m-linux_x86_64.whl 
下載以後，將whl文件重命名爲tensorflow-1.0.0-py3-none-linux_x86_64.whl，不然會出現

tensorflow-1.0.0-cp36-cp36m-linux_x86_64.whl is not a supported wheel on this platform.

同樣的報錯，具體參考https://github.com/tensorflow/tensorflow/issues/1990 
而後進入環境並安裝tensorflow

source activate tensorflow #激活tensorflow環境 cd /Downloads #切換到whl文件所在文件夾 pip install --ignore-installed --upgrade tensorflow-1.0.0-py3-none-linux_x86_64.whl #切記，不要用sudo pip，也不要用pip3，而後--ignore-installed --upgrade等參數也不能省略，不然會出錯。

 

3，安裝tensorflow(gpu版）

建立環境的命令爲：conda create -n tensorflow-gpu python=3.6 
先下載安裝包，下載路徑爲：https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.0.0-cp36-cp36m-linux_x86_64.whl 
而後進入環境並安裝tensorflow-gpu

source activate tensorflow-gpu #激活tensorflow環境 cd /Downloads #切換到whl文件所在文件夾 pip install --ignore-installed --upgrade tensorflow_gpu-1.0.0-cp36-cp36m-linux_x86_64.whl #切記，不要用sudo pip，也不要用pip3，而後--ignore-installed --upgrade等參數也不能省略，不然會出錯。

接着，還要配置cuda和cudnn，能夠到nvidia官網下載，接下來的配置可參考http://blog.csdn.net/jteng/article/details/52975247

4，驗證安裝

成功。

(tensorflow)$ python
import tensorflow as tf hello = tf.constant('Hello, TensorFlow!') sess = tf.Session() sess.run(hello)

 

存在的問題，運行時，兩個版本均有warning, NOT error，可是不影響結果，只是執行速度比較慢，聽說是由於爲了避免同框架上的可遷移性，尚未對cpu進行編譯，他建議你爲了更快的速度，能夠從編碼編譯，執行速度會更快。參考https://github.com/tensorflow/tensorflow/issues/8037

W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE3 instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.轉自：http://blog.csdn.net/michaelliang12/article/details/60106686