mnist_cnn.pyを動かしてみた

サンプルプログラムはここからダウンロードした。
https://github.com/keras-team/keras/blob/master/examples/mnist_cnn.py

さっそく実行。

Windows PowerShell
Copyright (C) Microsoft Corporation. All rights reserved.

PS > python mnist_cnn.py
Using TensorFlow backend.
x_train shape: (60000, 28, 28, 1)
60000 train samples
10000 test samples
Train on 60000 samples, validate on 10000 samples
Epoch 1/12
2018-02-28 14:28:39.205108: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\platform\cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to
 use: AVX AVX2
2018-02-28 14:28:39.467982: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\gpu\gpu_device.cc:1105] Found device 0 with properties:
name: GeForce GTX 970 major: 5 minor: 2 memoryClockRate(GHz): 1.2155
pciBusID: 0000:01:00.0
totalMemory: 4.00GiB freeMemory: 3.31GiB
2018-02-28 14:28:39.472806: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\gpu\gpu_device.cc:1195] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce GT
X 970, pci bus id: 0000:01:00.0, compute capability: 5.2)
60000/60000 [==============================] - 11s 176us/step - loss: 0.2587 - acc: 0.9195 - val_loss: 0.0584 - val_acc: 0.9808
Epoch 2/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0858 - acc: 0.9739 - val_loss: 0.0386 - val_acc: 0.9863
Epoch 3/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0654 - acc: 0.9807 - val_loss: 0.0317 - val_acc: 0.9886
Epoch 4/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0549 - acc: 0.9834 - val_loss: 0.0317 - val_acc: 0.9888
Epoch 5/12
60000/60000 [==============================] - 9s 143us/step - loss: 0.0476 - acc: 0.9860 - val_loss: 0.0311 - val_acc: 0.9894
Epoch 6/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0426 - acc: 0.9873 - val_loss: 0.0327 - val_acc: 0.9906
Epoch 7/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0385 - acc: 0.9882 - val_loss: 0.0278 - val_acc: 0.9906
Epoch 8/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0355 - acc: 0.9890 - val_loss: 0.0245 - val_acc: 0.9918
Epoch 9/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0326 - acc: 0.9897 - val_loss: 0.0284 - val_acc: 0.9904
Epoch 10/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0283 - acc: 0.9910 - val_loss: 0.0293 - val_acc: 0.9906
Epoch 11/12
60000/60000 [==============================] - 9s 142us/step - loss: 0.0283 - acc: 0.9914 - val_loss: 0.0270 - val_acc: 0.9919
Epoch 12/12
60000/60000 [==============================] - 8s 142us/step - loss: 0.0274 - acc: 0.9914 - val_loss: 0.0269 - val_acc: 0.9910
Test loss: 0.026897845212663196
Test accuracy: 0.991

正答率は99.1%かぁ。。。
プログラム動作中のGPU負荷は70%前後だった。
Core i5環境なんで多くを求めちゃいけないので、まあこんな感じかと。