anacondaは使わない。AWSGPU+Ubuntu14.04+jupyter+theano+chainer+OpenCV3.1.0+cuDNNな環境構築


環境

AWSのGPUインスタンスを用いる
・インスタンスタイプ: g2.2xlarge
・AMI ID: ubuntu/images/hvm-ssd/ubuntu-trusty-14.04-amd64-server-20160627 (ami-2d39803a)

ベースインストール

$ sudo apt-get update
$ sudo apt-get install build-essential cmake pkg-config
$ sudo apt-get install libjpeg-dev libtiff5-dev libjasper-dev libpng12-dev
$ sudo apt-get install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev
$ sudo apt-get install libxvidcore-dev libx264-dev
$ sudo apt-get install libgtk2.0-dev
$ sudo apt-get install libatlas-base-dev gfortran
$ sudo apt-get -y install cmake git libgtk2.0-dev ocl-icd-opencl-dev
$ python --version
Python 2.7.6
$ sudo apt-get install python-dev python-pip
$ sudo apt-get install unzip
$ pip install --upgrade pip --user
$ pip install ipython pyzmq tornado --user
$ pip install jsonschema --user
$ pip install numpy matplotlib scipy --user

cudaインストール

$ wget http://developer.download.nvidia.com/compute/cuda/7.5/Prod/local_installers/cuda-repo-ubuntu1404-7-5-local_7.5-18_amd64.deb
$ sudo dpkg -i cuda-repo-ubuntu1404-7-5-local_7.5-18_amd64.deb
$ sudo apt-get update
$ sudo apt-get install cuda
$ echo 'export PATH=/usr/local/cuda-7.5/bin:$PATH' >> ~/.bashrc
$ echo 'export LD_LIBRARY_PATH=/usr/local/cuda-7.5/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
$ source .bashrc
$ sudo init 6
$ cat /proc/driver/nvidia/version
NVRM version: NVIDIA UNIX x86_64 Kernel Module  367.57  Mon Oct  3 20:37:01 PDT 2016
GCC version:  gcc version 4.8.4 (Ubuntu 4.8.4-2ubuntu1~14.04.3) 
$ dpkg -l | grep nvidia
ii  nvidia-352                                367.57-0ubuntu0.14.04.1                             amd64        Transitional package for nvidia-367
ii  nvidia-352-dev                            367.57-0ubuntu0.14.04.1                             amd64        Transitional package for nvidia-367-dev
ii  nvidia-352-uvm                            352.39-0ubuntu1                                     amd64        Transitional package for nvidia-352
ii  nvidia-367                                367.57-0ubuntu0.14.04.1                             amd64        NVIDIA binary driver - version 367.57
ii  nvidia-367-dev                            367.57-0ubuntu0.14.04.1                             amd64        NVIDIA binary Xorg driver development files
ii  nvidia-modprobe                           352.39-0ubuntu1                                     amd64        Load the NVIDIA kernel driver and create device files
ii  nvidia-opencl-icd-352                     367.57-0ubuntu0.14.04.1                             amd64        Transitional package for nvidia-opencl-icd-367
ii  nvidia-opencl-icd-367                     367.57-0ubuntu0.14.04.1                             amd64        NVIDIA OpenCL ICD
ii  nvidia-prime                              0.6.2                                               amd64        Tools to enable NVIDIA's Prime
ii  nvidia-settings                           352.39-0ubuntu1                                     amd64        Tool for configuring the NVIDIA graphics driver

cudnnインストール

  • cuda-repo-ubuntu1404-7-5-local_7.5-18_amd64.deb の入手方法
    https://developer.nvidia.com/cudnn から登録して、Downloadボタンを押す。 Download cuDNN v4 (Feb 10, 2016), for CUDA 7.0 and later. を選択、さらに cuDNN v4 Library for Linux を選択してダウンロードしてscpで /home/ubuntu にファイル転送しておく。
$ cd ~
$ tar -xvzf cudnn-7.0-linux-x64-v4.0-prod.tgz
libcudnn.so libcudnn.so.4 libcudnn.so.4.0.7 libcudnn_static.aが入ってるディレクトリの絶対パスをLD_LIBRARY_PATHに追加
$ cd cuda/lib64
$ ls
libcudnn.so  libcudnn.so.4  libcudnn.so.4.0.7  libcudnn_static.a
$ pwd
/home/ubuntu/cuda/lib64
$ LD_LIBRARY_PATH=/home/ubuntu/cuda/lib64; export LD_LIBRARY_PATH
$ cd ~
$ sudo cp cuda/include/cudnn.h /usr/local/cuda/include
$ sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
$ sudo chmod a+r /usr/local/cuda/lib64/libcudnn*
$ export PATH=/usr/local/cuda/bin:$PATH
$ export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
$ source .bashrc

Jupyterインストール

$ sudo apt-get remove --purge ipython
$ sudo apt-get install libjpeg tk-dev
$ pip install jupyter --user
$ pip install wcwidth --user
$ ~/.local/bin/jupyter notebook --generate-config
$ vim ~/.jupyter/jupyter_notebook_config.py
$ python
>>> from notebook.auth import passwd
>>> passwd()
Enter password: 【なんらかのパスワードを入力】
Verify password: 【なんらかのパスワードを入力】
'sha1:84acf9ab4e38:5498f3245ea50214fd5bdd3aaa2050b33d5c5f91'
>>> exit()
$ vi ~/.jupyter/jupyter_notebook_config.py
+c.NotebookApp.ip ='*'
+c.NotebookApp.port = 8888
+c.NotebookApp.password = u'sha1:84acf9ab4e38:5498f3245ea50214fd5bdd3aaa2050b33d5c5f91'
$ echo 'export PATH=$PATH:~/.local/bin' >> ~/.bashrc
$ source ~/.bashrc

cudaパス関連付け

$ echo 'export PATH=/usr/local/cuda-7.5/bin:$PATH' >> ~/.bashrc
$ echo 'export LD_LIBRARY_PATH=/usr/local/cuda-7.5/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
$ source .bashrc

theano & chainerインストール

※ chainerのインストールをcudnnのインストールより先にやるとGPU処理が正常に通らなくなるので注意

$ pip install theano --user
$ pip install chainer --user

opencv3.1.0インストール

$ wget --no-check-certificate https://github.com/Itseez/opencv/archive/3.1.0.zip -O opencv-3.1.0.zip
$ unzip opencv-3.1.0.zip
$ cd opencv-3.1.0
$ git clone --depth 1 https://github.com/Itseez/opencv_contrib.git opencv_contrib
$ cd opencv_contrib
$ git fetch origin --tags --depth 1
$ git checkout 3.1.0
$ sudo apt-get -y -qq install cmake git libgtk2.0-dev ocl-icd-opencl-dev qt5-default
$ sudo apt-get install libhdf5-dev
$ pip install h5py --user
$ cd ..
$ mkdir build
$ cd build
$ cmake -D CMAKE_BUILD_TYPE=RelWithDebugInfo -D CMAKE_INSTALL_PREFIX=/usr/local -D WITH_TBB=ON -D OPENCV_EXTRA_MODULES_PATH=../opencv_contrib/modules -D BUILD_NEW_PYTHON_SUPPORT=ON -D WITH_V4L=ON -D INSTALL_C_EXAMPLES=ON  -D INSTALL_PYTHON_EXAMPLES=ON -D WITH_CUDA=ON -D ENABLE_PRECOMPILED_HEADERS=OFF -D CUDA_ARCH_BIN="3.0" -D CUDA_ARCH_PTX="" -D WITH_OPENGL=ON -D BUILD_EXAMPLES=ON -D WITH_QT=ON -D WITH_OPENGL=ON -D ENABLE_FAST_MATH=1 -D CUDA_FAST_MATH=1 ..
$ make -j 4
$ sudo make install -j 4
$ sudo ldconfig
$ echo 'export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
$ source ~/.bashrc
$ pkg-config --modversion opencv
3.1.0

OpenCVテスト

$ sudo ln /dev/null /dev/raw1394
$ cd ~
$ mkdir test
$ cd test
$ wget http://www.sonicjapan.co.jp/sample/download/AVI-352x240-2sec.AVI
$ python
>>> import numpy as np
>>> import cv2
>>> cap = cv2.VideoCapture('./AVI-352x240-2sec.AVI')
>>> cap.read()

jupyter notebookでの動画読込確認

$ cd ~
$ jupyter notebook
先ほどのtestディレクトリへ移動してnotebookを作り、下記実行
import numpy as np
import cv2
cap = cv2.VideoCapture('./AVI-352x240-2sec.AVI')
cap.read()

theanoのGPU処理テスト

$ cd ~
$ git clone https://github.com/aidiary/deep-learning-theano
$ cd ~/deep-learning-theano/scripts
$ THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32 python convnet.py
Using gpu device 0: GRID K520
....
Optimization complete.
Best validation score of 0.910000 % obtained at iteration 16400, with test performance 0.930000 %
The code for file convnet.py ran for 39.85m

chainerのGPU処理テスト

$ git clone https://github.com/pfnet/chainer.git
$ python ~/chainer/examples/mnist/train_mnist.py -g 0