텐서플로우 GPU 버전을 설치하면 최고인데, CPU 버전만 사용해야 한다. CPU 버전을 쓰면 아래와 같이 메세지가 나온다.
(tensorflow) now0930@:cifar10$ python cifar10_testv1.py Using TensorFlow backend. Couldn't import dot_parser, loading of dot files will not be possible. _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= conv2d_1 (Conv2D) (None, 32, 32, 32) 896 _________________________________________________________________ dropout_1 (Dropout) (None, 32, 32, 32) 0 _________________________________________________________________ conv2d_2 (Conv2D) (None, 32, 32, 32) 9248 _________________________________________________________________ max_pooling2d_1 (MaxPooling2 (None, 32, 16, 16) 0 _________________________________________________________________ flatten_1 (Flatten) (None, 8192) 0 _________________________________________________________________ dense_1 (Dense) (None, 512) 4194816 _________________________________________________________________ dropout_2 (Dropout) (None, 512) 0 _________________________________________________________________ dense_2 (Dense) (None, 10) 5130 ================================================================= Total params: 4,210,090 Trainable params: 4,210,090 Non-trainable params: 0 _________________________________________________________________ None Train on 50000 samples, validate on 10000 samples Epoch 1/25 2017-06-17 08:55:41.662236: 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. 2017-06-17 08:55:41.663359: 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. 2017-06-17 08:55:41.663384: 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. 288/50000 [..............................] - ETA: 707s - loss: 2.3271 - acc: 0.1076
연산 시간이 CPU 최신 기술로 줄어든다는데, 전에 해보렸으나 적당한 tutorial이 없어 못했다. 그러던 중, 아래 사이트를 찾았다.
“텐서플로우, 직접 소스코드 빌디하여 설치하기”
정리하면..
- Git에서 clone으로 다운로드
$ git clone https://github.com/tensorflow/tensorflow $ cd tensorflow $ git checkout r1.2 #r은 release..적용 버전이 1.2
- tensorflow을 컴파일하기 위한 Bazel 설치
세가지 방법으로 Bazel 설치 가능하다. 나는 debian을 사용하므로, compile 했다.- Install required Package
apt-get install pkg-config zip g++ zlib1g-dev unzip
- Download..
- 실행 권한을 주고 run
- bazel 경로를 PATH 변수에 추가
export PATH="$PATH:$HOME/bin"
- Install required Package
- tensorflow 컴파일 옵션 설정..configure..
#아까 다운로드 받은 tensorflow 디렉토리에서.. ./configure
- Bazel로 빌드하기
bazel build --config=opt //tensorflow/tools/pip_package:build_pip_package
–config=opt 옵션이 아까 경고 부분을 해결한다.
- Pip 설치용 whl 만들기
$./bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
bazel-bin이 링크다. 경로를 잘 확인한다. 패키지 파일을 /tmp 디렉토리에 넣어야 되는지 모르겠다.
- conda 가상 환경 활성화..
- 기존 tensorflow 설치버전 무시하고 설치
(tensorflow)$ pip install --ignore-installed --upgrade tfBinaryURL
tfBinaryURL이 아까 만든 /tmp/tensorflow_pkg의 whl 경로이다.
설치 후, python을 다시 실행하면 해당 메세지가 없어졌다.
(tensorflow) now0930@:cifar10$ python cifar10_testv1.py Using TensorFlow backend. Couldn't import dot_parser, loading of dot files will not be possible. _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= conv2d_1 (Conv2D) (None, 32, 32, 32) 896 _________________________________________________________________ dropout_1 (Dropout) (None, 32, 32, 32) 0 _________________________________________________________________ conv2d_2 (Conv2D) (None, 32, 32, 32) 9248 _________________________________________________________________ max_pooling2d_1 (MaxPooling2 (None, 32, 16, 16) 0 _________________________________________________________________ flatten_1 (Flatten) (None, 8192) 0 _________________________________________________________________ dense_1 (Dense) (None, 512) 4194816 _________________________________________________________________ dropout_2 (Dropout) (None, 512) 0 _________________________________________________________________ dense_2 (Dense) (None, 10) 5130 ================================================================= Total params: 4,210,090 Trainable params: 4,210,090 Non-trainable params: 0 _________________________________________________________________ None Train on 50000 samples, validate on 10000 samples Epoch 1/25
설치하기 전에는 속도가 많이 빨라질 것이라 기대 했으나..별 차이 없어보인다.
