MacBookにAMDのGPUが搭載されていたのでPlaidMLを使って画像分類を試してみました。
PlaidMLのGitHub： https://github.com/plaidml/plaidml
【環境】
PC: MacBook Pro (15-inch, 2017)
OS: macOS High Sierra(10.13.3)
GPU: Radeon Pro 560 (4GB)
Python3(virtualenv)
まずは手持ちのMacでOpenCLが使えるか確認するためにbrewでclinfoをインストールします
$ brew install clinfo
確認
$ clinfo
大丈夫そうですね
Number of platforms 1
AMDのGPUを発見
clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...) Apple
clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...) Success [P0]
clCreateContext(NULL, ...) [default] Success [P0]
clCreateContextFromType(NULL, CL_DEVICE_TYPE_DEFAULT) Success (2)
Platform Name Apple
Device Name Intel(R) HD Graphics 630
Device Name AMD Radeon Pro 560 Compute Engine
clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU) Success (1)
Platform Name Apple
Device Name Intel(R) Core(TM) i7-7920HQ CPU @ 3.10GHz
clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU) Success (2)
Platform Name Apple
Device Name Intel(R) HD Graphics 630
Device Name AMD Radeon Pro 560 Compute Engine
clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR) No devices found in platform
clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM) <checkNullCtxFromType:2437: create context from type CL_DEVICE_TYPE_CUSTOM : error -30>
clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL) Success (3)
Platform Name Apple
Device Name Intel(R) HD Graphics 630
Device Name AMD Radeon Pro 560 Compute Engine
Device Name Intel(R) Core(TM) i7-7920HQ CPU @ 3.10GHz
Pythonの環境を用意
$ mkvirtualenv plaidmlPy3 -p python3
Running virtualenv with interpreter /usr/local/bin/python3
Using base prefix '/usr/local/Cellar/python/3.6.5/Frameworks/Python.framework/Versions/3.6'
New python executable in /Users/ユーザー名/.virtualenvs/plaidmlPy3/bin/python3.6
Also creating executable in /Users/ユーザー名/.virtualenvs/plaidmlPy3/bin/python
Installing setuptools, pip, wheel...done.
virtualenvwrapper.user_scripts creating /Users/ユーザー名/.virtualenvs/plaidmlPy3/bin/predeactivate
virtualenvwrapper.user_scripts creating /Users/ユーザー名/.virtualenvs/plaidmlPy3/bin/postdeactivate
virtualenvwrapper.user_scripts creating /Users/ユーザー名/.virtualenvs/plaidmlPy3/bin/preactivate
virtualenvwrapper.user_scripts creating /Users/ユーザー名/.virtualenvs/plaidmlPy3/bin/postactivate
virtualenvwrapper.user_scripts creating /Users/ユーザー名/.virtualenvs/plaidmlPy3/bin/get_env_details
PlaidMLをインストール
$ pip install plaidml-keras
Collecting plaidml-keras
Downloading plaidml_keras-0.3.1-py2.py3-none-any.whl
Collecting six (from plaidml-keras)
Using cached six-1.11.0-py2.py3-none-any.whl
Collecting keras==2.0.8 (from plaidml-keras)
Downloading Keras-2.0.8-py2.py3-none-any.whl (276kB)
100% |████████████████████████████████| 276kB 615kB/s
Collecting plaidml (from plaidml-keras)
Downloading plaidml-0.3.1-py2.py3-none-macosx_10_10_x86_64.whl (14.1MB)
100% |████████████████████████████████| 14.1MB 113kB/s
Collecting pyyaml (from keras==2.0.8->plaidml-keras)
Using cached PyYAML-3.12.tar.gz
Collecting scipy>=0.14 (from keras==2.0.8->plaidml-keras)
Downloading scipy-1.0.1-cp36-cp36m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl (16.4MB)
100% |████████████████████████████████| 16.4MB 95kB/s
Requirement already satisfied: numpy>=1.9.1 in ./.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from keras==2.0.8->plaidml-keras)
Collecting enum34>=1.1.6 (from plaidml->plaidml-keras)
Downloading enum34-1.1.6-py3-none-any.whl
Collecting requests (from plaidml->plaidml-keras)
Downloading requests-2.18.4-py2.py3-none-any.whl (88kB)
100% |████████████████████████████████| 92kB 1.2MB/s
Collecting idna<2.7,>=2.5 (from requests->plaidml->plaidml-keras)
Downloading idna-2.6-py2.py3-none-any.whl (56kB)
100% |████████████████████████████████| 61kB 2.0MB/s
Collecting urllib3<1.23,>=1.21.1 (from requests->plaidml->plaidml-keras)
Downloading urllib3-1.22-py2.py3-none-any.whl (132kB)
100% |████████████████████████████████| 133kB 150kB/s
Collecting certifi>=2017.4.17 (from requests->plaidml->plaidml-keras)
Downloading certifi-2018.1.18-py2.py3-none-any.whl (151kB)
100% |████████████████████████████████| 153kB 1.1MB/s
Collecting chardet<3.1.0,>=3.0.2 (from requests->plaidml->plaidml-keras)
Downloading chardet-3.0.4-py2.py3-none-any.whl (133kB)
100% |████████████████████████████████| 143kB 1.2MB/s
Building wheels for collected packages: pyyaml
Running setup.py bdist_wheel for pyyaml ... done
Stored in directory: /Users/ユーザー名/Library/Caches/pip/wheels/2c/f7/79/13f3a12cd723892437c0cfbde1230ab4d82947ff7b3839a4fc
Successfully built pyyaml
Installing collected packages: six, pyyaml, scipy, keras, enum34, idna, urllib3, certifi, chardet, requests, plaidml, plaidml-keras
Successfully installed certifi-2018.1.18 chardet-3.0.4 enum34-1.1.6 idna-2.6 keras-2.0.8 plaidml-0.3.1 plaidml-keras-0.3.1 pyyaml-3.12 requests-2.18.4 scipy-1.0.1 six-1.11.0 urllib3-1.22
PlaidMLの環境をセットアップしていきます
$ plaidml-setup
「y」をタイプ
PlaidML Setup (0.3.1)
Thanks for using PlaidML!
Some Notes:
* Bugs and other issues: https://github.com/plaidml/plaidml
* Questions: https://stackoverflow.com/questions/tagged/plaidml
* Say hello: https://groups.google.com/forum/#!forum/plaidml-dev
* PlaidML is licensed under the GNU AGPLv3
Default Config Devices:
No devices.
Experimental Config Devices:
llvm_preview_cpu.0 : LLVM_preview_CPU
amd_radeon_pro_560_compute_engine.0 : AMD AMD Radeon Pro 560 Compute Engine
intel(r)_hd_graphics_630.0 : Intel Inc. Intel(R) HD Graphics 630
opencl_cpu.0 : Intel OpenCL CPU
Using experimental devices can cause poor performance, crashes, and other nastiness.
Enable experimental device support? (y,n)[n]:y
PlaidMLで使用できるデバイスIDが表示されます。
AMDのRADEON PRO 560を使用したいので、そのインデックス番号の「2」をタイプ
Multiple devices detected (You can override by setting PLAIDML_DEVICE_IDS).
Please choose a default device:
1 : llvm_preview_cpu.0
2 : amd_radeon_pro_560_compute_engine.0
3 : intel(r)_hd_graphics_630.0
4 : opencl_cpu.0
Default device? (1,2,3,4)[1]:2
PlaidMLを応援したいので「y」をタイプ! (別にレポートしたくない方は「n」を選択)
Selected device:
amd_radeon_pro_560_compute_engine.0
PlaidML sends anonymous usage statistics to help guide improvements.
We'd love your help making it better.
Enable telemetry reporting? (y,n)[y]:y
/Users/ユーザー名/.plaidmlに設定を保存したいので「y」をタイプ
Almost done. Multiplying some matrices...
Tile code:
function (B[X,Z], C[Z,Y]) -> (A) { A[x,y : X,Y] = +(B[x,z] * C[z,y]); }
Whew. That worked.
Save settings to /Users/ユーザー名/.plaidml? (y,n)[y]:y
Success!
サンプルを動かしたいのでクローンします
$ git clone https://github.com/plaidml/plaidbench.git
Cloning into 'plaidbench'...
remote: Counting objects: 296, done.
remote: Compressing objects: 100% (7/7), done.
remote: Total 296 (delta 2), reused 5 (delta 1), pack-reused 287
Receiving objects: 100% (296/296), 9.68 MiB | 757.00 KiB/s, done.
Resolving deltas: 100% (141/141), done.
移動
$ cd plaidbench
必要なモジュールをインストールする前に一応内容を確認
$ cat requirements.txt
fmfm
h5py>=2.7.0
plaidml-keras>=0.3.0
onnx-plaidml>=0.3.0
インストール開始
$ pip install -r requirements.txt
Collecting h5py>=2.7.0 (from -r requirements.txt (line 1))
Downloading h5py-2.7.1-cp36-cp36m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl (4.7MB)
100% |████████████████████████████████| 4.8MB 288kB/s
Requirement already satisfied: plaidml-keras>=0.3.0 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from -r requirements.txt (line 2))
Collecting onnx-plaidml>=0.3.0 (from -r requirements.txt (line 3))
Downloading onnx_plaidml-0.3.1-py2.py3-none-any.whl
Requirement already satisfied: six in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from h5py>=2.7.0->-r requirements.txt (line 1))
Requirement already satisfied: numpy>=1.7 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from h5py>=2.7.0->-r requirements.txt (line 1))
Requirement already satisfied: keras==2.0.8 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Requirement already satisfied: plaidml in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Collecting onnx>=1.1.0 (from onnx-plaidml>=0.3.0->-r requirements.txt (line 3))
Downloading onnx-1.1.1.tar.gz (940kB)
100% |████████████████████████████████| 942kB 1.1MB/s
Collecting click (from onnx-plaidml>=0.3.0->-r requirements.txt (line 3))
Downloading click-6.7-py2.py3-none-any.whl (71kB)
100% |████████████████████████████████| 71kB 20kB/s
Requirement already satisfied: pyyaml in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from keras==2.0.8->plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Requirement already satisfied: scipy>=0.14 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from keras==2.0.8->plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Requirement already satisfied: enum34>=1.1.6 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from plaidml->plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Requirement already satisfied: requests in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from plaidml->plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Collecting protobuf (from onnx>=1.1.0->onnx-plaidml>=0.3.0->-r requirements.txt (line 3))
Using cached protobuf-3.5.2.post1-cp36-cp36m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Requirement already satisfied: chardet<3.1.0,>=3.0.2 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from requests->plaidml->plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Requirement already satisfied: urllib3<1.23,>=1.21.1 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from requests->plaidml->plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Requirement already satisfied: idna<2.7,>=2.5 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from requests->plaidml->plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Requirement already satisfied: certifi>=2017.4.17 in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from requests->plaidml->plaidml-keras>=0.3.0->-r requirements.txt (line 2))
Requirement already satisfied: setuptools in /Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages (from protobuf->onnx>=1.1.0->onnx-plaidml>=0.3.0->-r requirements.txt (line 3))
Building wheels for collected packages: onnx
Running setup.py bdist_wheel for onnx ... done
Stored in directory: /Users/ユーザー名/Library/Caches/pip/wheels/40/c3/23/6190428e9a6f136bc0b3f8181b0eda5fcb56124c40b6d38878
Successfully built onnx
Installing collected packages: h5py, protobuf, onnx, click, onnx-plaidml
Successfully installed click-6.7 h5py-2.7.1 onnx-1.1.1 onnx-plaidml-0.3.1 protobuf-3.5.2.post1
ちゃんとインストールされている確認するためにMobileNetを起動してみましょう
$ python plaidbench.py mobilenet
大丈夫そう！
/Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
from ._conv import register_converters as _register_converters
Running 1024 examples with mobilenet, batch size 1
INFO:plaidml:Opening device "amd_radeon_pro_560_compute_engine.0"
Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.6/mobilenet_1_0_224_tf.h5
16752640/17225924 [============================>.] - ETA: 0sModel loaded.
Example finished, elapsed: 1.9480268955230713 (compile), 9.947688102722168 (execution), 0.009714539162814617 (execution per example)
Correctness: PASS, max_error: 1.9290237105451524e-05, max_abs_error: 1.0132789611816406e-06, fail_ratio: 0.0
ではでは本題の画像分類に取り掛かりましょう
plaidvisionをクローン
git clone https://github.com/plaidml/plaidvision.git
移動
cd plaidvision/
インストールする内容を確認
$ cat requirements.txt
h5py>=2.7.0
imageio>=2.2.0
Pillow>=4.2.1
pygame>=1.9.3
plaidml-keras>=0.1.0a5
opencv-python>=3.3.0.10
実行
$ python plaidvision.py mobilenet
別ウィンドウが起動します！
Using PlaidML backend.
/Users/ユーザー名/.virtualenvs/plaidmlPy3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
from ._conv import register_converters as _register_converters
INFO:plaidml:Opening device "amd_radeon_pro_560_compute_engine.0"
その画面をスクショしてみました！
ちゃんとwater_bottleと認識されていますね。
先ほどmobilenetを指定しましたが他にinception_v3, resnet50, vgg16, vgg19, xceptionのモデルもあるようです。
$ python plaidvision.py xception
CUDAのセットアップを経験した人なら分かると思いますが、非常に簡単にセットアップできた思います。
あと、kerasのバックエンドの置き換えはこれだけみたいです。
import plaidml.keras
plaidml.keras.install_backend()
NVIDIAのGPUじゃなくても機械学習捗りそうですね！
プライベードで画像を集めてkerasを使って作成したデモがあるのでPlaidMLで動いたら別の機会に紹介します。
【ネタ編】
オワカリイタダケタダロウカ？
モウイチド...
だれがトイプードルじゃ！