介绍Google的TensorFlow机器学习开源库,在UbuntuKylin上的安装和和源码编译。
原始官方文档参见:http://www.tensorflow.org.
本电脑配置如下:
3.19.0-15-generic #15-Ubuntu x86_64 GNU/Linux
NVIDIA Corporation GK110BGL [Tesla K40c]
NVIDIA Corporation GK110GL [Quadro K5200]
Python 2.7
Cuda toolkit = 7.5
cuDNN = 7.5 v5
gcc = 4.9
g++ = 4.9
Bazel = 0.4.4
TensorFlow学习资源推荐
tensorflow中文入门教程-含视频
tensorflow入门视频教程-含互动
tensorflow中文社区
TensorFlow 官方文档中文版
TensorFlow在图像识别中的应用
本文是在安装caffe之后,继续安装TensorFlow,下面有些CUDA和 CUDNN的安装可见 Caffe + Ubuntu 15.04 + CUDA 7.5 在服务器上安装配置及卸载重新安装(已测试可执行)
安装TensorFlow的Requirements
Python 2.7 and Python 3.3+
Cuda toolkit >= 7.0
cuDNN >= v3
gcc > 4.8
g++ > 4.8
Bazel > 0.4.2
一、安装依赖包
1. 安装Tensorflow python API
sudo apt-get install python-pip python-dev
sudo apt-get install python-numpy swig python-dev
sudo apt-get install Git
2. 安装 Bazel
TensorFlow Serving requires Bazel 0.4.2 or higher,Bazel的安装可见官网。
OpenJDK做为GPL许可(GPL-licensed)的Java平台的开源化实现,Sun正式发布它已经六年有余。从发布那一时刻起,Java社区的大众们就又开始努力学习,以适应这个新的开源代码基础(code-base)。 [1]
OpenJDK在2013年发展迅速,被著名IT杂志SD Times评选为2013 SD Times 100,位于“极大影响力”分类第9位。
http://www.infoq.com/cn/news/2015/03/google-open-source-bazel
Google日前开源了他们内部使用的构建工具Bazel。
Bazel是一个类似于Make的工具,是Google为其内部软件开发的特点量身定制的工具,如今Google使用它来构建内部大多数的软件。它的功能有诸多亮点:
多语言支持:目前Bazel默认支持Java、Objective-C和C++,但可以被扩展到其他任何变成语言。
高级构建描述语言:项目是使用一种叫BUILD的语言来描述的,它是一种简洁的文本语言,它把一个项目视为一个集合,这个集合由一些互相关联的库、二进制文件和测试用例组成。相反,像Make这样的工具,需要去描述每个文件如何调用编译器。
多平台支持:同一套工具和相同的BUILD文件可以用来为不同的体系结构构建软件,甚至是不同的平台。在Google,Bazel被同时用在数据中心系统中的服务器应用和手机端的移动应用上。
可重复性:在BUILD文件中,每个库、测试用例和二进制文件都需要明确指定它们的依赖关系。当一个源码文件被修改时,Bazel凭这些依赖来判断哪些部分需要重新构建,以及哪些任务可以并行进行。这意味着所有构建都是增量的,并且相同构建总是产生一样的结果。
可伸缩性:Bazel可以处理大型项目;在Google,一个服务器软件有十万行代码是很常见的,在什么都不改的前提下重新构建这样一个项目,大概只需要200毫秒。
JDK8的安装(必须的)
sudo apt-get install openjdk-8-jdk openjdk-8-source
sudo apt-get install pkg-config zip g++ zlib1g-dev unzip
sudo add-apt-repository ppa:webupd8team/java #添加仓库
sudo apt-get update #更新软件列表
sudo apt-get install oracle-java8-installer #正式安装jdk8
java -version # 验证安装
2.1 安装 Bazel-方法1
echo “deb http://storage.googleapis.com/bazel-apt stable jdk1.8” | sudo tee /etc/apt/sources.list.d/bazel.list
curl https://storage.googleapis.com/bazel-apt/doc/apt-key.pub.gpg | sudo apt-key add -
sudo apt-get update
sudo apt-get install bazel
sudo apt-get upgrade bazel
bazel version
2.2 安装 Bazel-方法2
Bazel 下载链接
cd ~/Downloads
chmod +x bazel-0.4.5-installer-linux-x86_64.sh #对.sh文件授权
./bazel-0.4.5-installer-linux-x86_64.sh --user #运行.sh文件
bazel version
设置环境变量
export PATH="$PATH:$HOME/bin"
可能出现的问题
W: 无法下载 http://storage.googleapis.com/bazel-apt/dists/stable/InRelease Unable to find expected entry ‘jdk1.8/binary-i386/Packages’ in Release file (Wrong sources.list entry or malformed file)
E: Some index files failed to download. They have been ignored, or old ones used instead.
的错误
解决方法
sudo gedit /etc/apt/sources.list.d/bazel.list
将deb http://storage.googleapis.com/bazel-apt stable jdk1.8修改为deb [arch=amd64] http://storage.googleapis.com/bazel-apt stable jdk1.8
3. CUDA和 CUDNN的安装,在 Linux 上开启 GPU 支持
为了编译并运行能够使用 GPU 的 TensorFlow, 需要先安装 NVIDIA 提供的 Cuda Toolkit 7.5 和 CUDNN 7.5 V5
TensorFlow 的 GPU 特性只支持 NVidia Compute Capability >= 3.5 的显卡. 被支持的显卡 包括但不限于
NVidia Titan
NVidia Titan X
NVidia K20
NVidia K40
可见 Caffe + Ubuntu 15.04 + CUDA 7.5 在服务器上安装配置及卸载重新安装(已测试可执行)
二、Ubuntu/Linux直接安装
# 仅使用 CPU 的版本
$ pip install https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.0.1-cp27-none-linux_x86_64.whl
# 开启 GPU 支持的版本 (安装该版本的前提是已经安装了 CUDA sdk)
$ pip install https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-1.0.1-cp27-none-linux_x86_64.whl
三、源码编译
TensorFlow 源码安装官方教程
3.1 克隆 TensorFlow 仓库
git clone --recurse-submodules https://github.com/tensorflow/tensorflow #拉取源代码
–recurse-submodules 参数是必须得, 用于获取 TesorFlow 依赖的 protobuf 库
3.2 配置 TensorFlow 的 Cuba 选项
cd tensorflow
./configure # 配置tensorflow
执行configure的时候会问你问题
Please specify the location of python. [Default is /usr/bin/python]
Please specify optimization flags to use during compilation [Default is -march=native]
Do you wish to use jemalloc as the malloc implementation? [Y/N]
y
Do you wish to build TensorFlow with Google Cloud Platform support? [Y/N]
y
Do you wish to build TensorFlow with Hadoop File System support? [Y/N]
y
Do you wish to build TensorFlow with the XLA just-in-time compiler (experimental)? [Y/N]
y
Do you wish to build TensorFlow with OpenCL support? [Y/N]
n
Do you wish to build TensorFlow with CUDA support? [Y/N]
y
若 Do you wish to build TensorFlow with OpenCL support? [Y/N] 中选择 y,则需要安装 OpenCL drivers 和 ComputeCpp compiler,具体步骤可参考
Optional: Install OpenCL (Experimental, Linux only)
tensorflow-opencl
否则,会出现如下一直循环的情况。
3.3 编译
mkdir /tmp/tensorflow_pkg
3.3.1 仅 CPU 支持,无 GPU 支持
cd tensorflow
bazel build -c opt //tensorflow/tools/pip_package:build_pip_package
出现的问题
The 'build' command is only supported from within a workspace
解决方法
cd tensorflow
3.3.2 有 GPU 支持
cd tensorflow
bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
3.3.3 生成 pip安装包
bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
cd 到 /tmp/tensorflow_pkg目录下,找到编译好的whl文件
cd /tmp/tensorflow_pkg
sudo pip install --config=cuda tensorflow-1.0.1-cp27-none-linux_x86_64.whl
3.3.4 编译目标程序, 开启 GPU 支持
bazel build -c opt --config=cuda //tensorflow/cc:tutorials_example_trainer
bazel-bin/tensorflow/cc/tutorials_example_trainer --use_gpu
四、设置TensorFlow环境
cd tensorflow
bazel build -c opt //tensorflow/tools/pip_package:build_pip_package
# To build with GPU support:
bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
mkdir _python_build
cd _python_build
ln -s ../bazel-bin/tensorflow/tools/pip_package/build_pip_package.runfiles/org_tensorflow/* .
ln -s ../tensorflow/tools/pip_package/* .
sudo python setup.py develop
五、测试TensorFlow
import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))
Hello, TensorFlow!
a = tf.constant(10)
b = tf.constant(32)
print(sess.run(a+b))
42
用tensorflow实现梵高作画
1. neural-style下载在这个[github网站下载相应代码]
2. 下载vgg19
3. 将imagenet-vgg-verydeep-19.mat复制到neural-style的文件夹根目录下
cp -r imagenet-vgg-verydeep-19.mat /home/bids/neural-style-master/
4. 执行梵高作画
python neural_style.py –content ./example/xxx.jpg (此括号内不要复制:xxx代表你想要使用的图片名称) –styles ./example/ 1-style.jpg(此括号内不要复制:1-style.jpg是梵高星空图片在文件夹内名称) –output ./example/yyy.jpg (yyy代表你想要生成的图片名称)
cd neural-style-master
python neural_style.py –content ./example/1-content.jpg --styles ./example/1-style.jpg --output ./example/1-output.jpg
六、出现的问题
gcc 版本 -fno-canonical-system-headers
当执行
./configure
出现如下问题
INFO: Found 1 target...
Slow read: a 51765952-byte read from /home/bids/.cache/bazel/_bazel_bids/5df0e0fb624204ab1c5ce0472e695b94/external/local_config_cuda/cuda/lib/libcurand.so.7.5 took 9675ms.
INFO: From Compiling external/llvm/lib/Support/Host.cpp:
external/llvm/lib/Support/Host.cpp: In function 'llvm::StringRef llvm::sys::getHostCPUName()':
external/llvm/lib/Support/Host.cpp:898:5: warning: 'Type' may be used uninitialized in this function [-Wuninitialized]
external/llvm/lib/Support/Host.cpp:964:7: warning: 'Subtype' may be used uninitialized in this function [-Wmaybe-uninitialized]
ERROR: /home/bids/.cache/bazel/_bazel_bids/5df0e0fb624204ab1c5ce0472e695b94/external/llvm/BUILD:1667:1: C++ compilation of rule '@llvm//:support' failed: gcc failed: error executing command /usr/bin/gcc -U_FORTIFY_SOURCE -fstack-protector -Wall -B/usr/bin -B/usr/bin -Wunused-but-set-parameter -Wno-free-nonheap-object -fno-omit-frame-pointer -g0 -O2 '-D_FORTIFY_SOURCE=1' -DNDEBUG ... (remaining 43 argument(s) skipped): com.google.devtools.build.lib.shell.BadExitStatusException: Process exited with status 1.
In file included from external/llvm/lib/Support/DynamicLibrary.cpp:16:0:
external/llvm/include/llvm/ADT/DenseSet.h:226:16: error: 'using llvm::DenseSet::BaseT::BaseT' conflicts with a previous declaration
external/llvm/include/llvm/ADT/DenseSet.h:223:39: note: previous declaration 'using BaseT = class llvm::detail::DenseSetImpl >, ValueInfoT>'
external/llvm/include/llvm/ADT/DenseSet.h:244:16: error: 'using llvm::SmallDenseSet::BaseT::BaseT' conflicts with a previous declaration
external/llvm/include/llvm/ADT/DenseSet.h:241:18: note: previous declaration 'using BaseT = class llvm::detail::DenseSetImpl >, ValueInfoT>'
Target //tensorflow/tools/pip_package:build_pip_package failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 54.671s, Critical Path: 28.01s
bids@bids-HP-Z840-Workstation:~/tensorflow$ bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
WARNING: /home/bids/tensorflow/tensorflow/contrib/learn/BUILD:15:1: in py_library rule //tensorflow/contrib/learn:learn: target '//tensorflow/contrib/learn:learn' depends on deprecated target '//tensorflow/contrib/session_bundle:exporter': Use SavedModel Builder instead.
WARNING: /home/bids/tensorflow/tensorflow/contrib/learn/BUILD:15:1: in py_library rule //tensorflow/contrib/learn:learn: target '//tensorflow/contrib/learn:learn' depends on deprecated target '//tensorflow/contrib/session_bundle:gc': Use SavedModel instead.
INFO: Found 1 target...
ERROR: /home/bids/.cache/bazel/_bazel_bids/5df0e0fb624204ab1c5ce0472e695b94/external/zlib_archive/BUILD.bazel:5:1: C++ compilation of rule '@zlib_archive//:zlib' failed: crosstool_wrapper_driver_is_not_gcc failed: error executing command external/local_config_cuda/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc -U_FORTIFY_SOURCE '-D_FORTIFY_SOURCE=1' -fstack-protector -fPIE -Wall -Wunused-but-set-parameter ... (remaining 37 argument(s) skipped): com.google.devtools.build.lib.shell.BadExitStatusException: Process exited with status 1.
gcc: error: unrecognized command line option '-fno-canonical-system-headers'
Target //tensorflow/tools/pip_package:build_pip_package failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 4.726s, Critical Path: 1.88s
解决方法:
这是因为gcc 版本的问题。因之前安装caffe 所需的gcc版本为4.7,故升级到4.9版本即可。可参考
Porting to GCC 4.7
Caffe + Ubuntu 15.04 + CUDA 7.5 在服务器上安装配置及卸载重新安装(已测试可执行)
cd /usr/bin
sudo rm gcc
sudo ln -s gcc-4.9 gcc
sudo rm g++
sudo ln -s g++-4.9 g++
问题 Oracle JDK 8 is not installed
当执行如下
sudo apt-get install openjdk-8-jdk openjdk-8-source
出现如下错误
download failed
Oracle JDK 8 is NOT installed.
dpkg: error processing package oracle-java8-installer (--configure):
subprocess installed post-installation script returned error exit status 1
Errors were encountered while processing:
oracle-java8-installer
E: Sub-process /usr/bin/dpkg returned an error code (1)
解决方法: 这是因为oracle-java8-installer 不能下载或者下载不完整导致的。
手动下载,见链接。
cp -r jdk-8u121-linux-x64.tar.gz /var/cache/oracle-jdk8-installer/
sudo apt-get install oracle-jdk8-installer
问题 TensorFlow ImportError: cannot import name pywrap_tensorflow
当执行如下
cd tensorflow
import tensorflow as tf
出现如下错误
Traceback (most recent call last):
File "" , line 1, in <module>
File "tensorflow/__init__.py", line 23, in <module>
from tensorflow.Python import *
File "tensorflow/python/__init__.py", line 48, in <module>
from tensorflow.python import pywrap_tensorflow
ImportError: cannot import name pywrap_tensorflow
解决方法: 这是因为python误以为tensorflow目录中的tensorflow就是要导入的模块
不要在tensorflow中运行python或者ipython
更改keras的backend 设置 tensorflow,theano
sudo gedit ~/.keras/keras.json
Theano为后端
{
"image_dim_ordering": "th",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "theano"
}
Tensorflow为后端
{
"image_dim_ordering": "tf",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "tensorflow"
}
参考文献:
TensorFlow源码编译-基于Ubuntu 15.04
TensorFlow 研究实践 一
Ubuntu安装Bazel
官网教程 Installing Bazel
搭建Tensorflow虚拟机学习环境
TensorFlow的安装
TensorFlow 从入门到精通(一):安装和使用
ubuntu16.04下安装TensorFlow(GPU加速)—-详细图文教程
Ubuntu: Oracle JDK 8 is NOT installed
教你从头到尾利用DL学梵高作画:GTX 1070 cuda 8.0 tensorflow gpu版