只贴代码：
原始数据集下载：http://yann.lecun.com/exdb/mnist/
代码说明：http://keras.io/getting-started/sequential-model-guide/#examples
"""
@version:
@author: vinsin
@license: Apache Licence
@software: PyCharm
@file: test_keras.py
@time: 16-7-19 下午4:53
"""
def load_mnist(path, kind='train'):
"""Load MNIST data from `path`"""
labels_path = os.path.join(path, '%s-labels-idx1-ubyte' % kind)
images_path = os.path.join(path, '%s-images-idx3-ubyte' % kind)
with open(labels_path, 'rb') as lbpath:
magic, n = struct.unpack('>II', lbpath.read(8))
labels = np.fromfile(lbpath, dtype=np.uint8)
with open(images_path, 'rb') as imgpath:
magic, num, rows, cols = struct.unpack(">IIII", imgpath.read(16))
images = np.fromfile(imgpath, dtype=np.uint8).reshape(len(labels), 784)
return images, labels
X_train, y_train = load_mnist('../data', kind='train')
print('Rows: %d, columns: %d' % (X_train.shape[0], X_train.shape[1]))
X_test, y_test = load_mnist('../data', kind='t10k')
print('Rows: %d, columns: %d' % (X_test.shape[0], X_test.shape[1]))
import theano
theano.config.floatX = 'float32'
X_train = X_train.astype(theano.config.floatX)
X_test = X_test.astype(theano.config.floatX)
from keras.utils import np_utils
print('First 3 data: ', X_train[:3])
print('First 3 labels: ', y_train[:3])
y_train_ohe = np_utils.to_categorical(y_train)
print('First 3 labels (one-hot):', y_train_ohe[:3])
from keras.models import Sequential
from keras.layers.core import Dense
from keras.optimizers import SGD
np.random.seed(1)
model = Sequential()
model.add(Dense(input_dim=X_train.shape[1],
output_dim=50,
init='uniform',
activation='tanh'))
model.add(Dense(input_dim=50,
output_dim=50,
init='uniform',
activation='tanh'))
model.add(Dense(input_dim=50,
output_dim=y_train_ohe.shape[1],
init='uniform',
activation='softmax'))
sgd = SGD(lr=0.001, decay=1e-7, momentum=.9)
model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=["accuracy"])
model.fit(X_train,
y_train_ohe,
nb_epoch=50,
batch_size=300,
verbose=1,
validation_split=0.1)
y_train_pred = model.predict_classes(X_train, verbose=0)
print('First 3 predictions: ', y_train_pred[:3])
train_acc = np.sum(y_train == y_train_pred, axis=0) / X_train.shape[0]
print('Training accuracy: %.2f%%' % (train_acc * 100))
y_test_pred = model.predict_classes(X_test, verbose=0)
test_acc = np.sum(y_test == y_test_pred, axis=0) / X_test.shape[0]
print('Test accuracy: %.2f%%' % (test_acc * 100))
输出的日志信息如下：
Rows: 60000, columns: 784
Rows: 10000, columns: 784
Using Theano backend.
First 3 data: [[ 0. 0. 0. ..., 0. 0. 0.]
[ 0. 0. 0. ..., 0. 0. 0.]
[ 0. 0. 0. ..., 0. 0. 0.]]
First 3 labels: [5 0 4]
First 3 labels (one-hot): [[ 0. 0. 0. 0. 0. 1. 0. 0. 0. 0.]
[ 1. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 1. 0. 0. 0. 0. 0.]]
Train on 54000 samples, validate on 6000 samples
Epoch 1/50
54000/54000 [==============================] - 1s - loss: 2.2290 - acc: 0.3594 - val_loss: 2.1092 - val_acc: 0.5343
Epoch 2/50
54000/54000 [==============================] - 1s - loss: 1.8848 - acc: 0.5272 - val_loss: 1.6081 - val_acc: 0.5518
Epoch 3/50
54000/54000 [==============================] - 1s - loss: 1.3900 - acc: 0.5886 - val_loss: 1.1653 - val_acc: 0.6645
Epoch 4/50
54000/54000 [==============================] - 1s - loss: 1.0614 - acc: 0.6958 - val_loss: 0.9047 - val_acc: 0.7703
Epoch 5/50
54000/54000 [==============================] - 1s - loss: 0.8566 - acc: 0.7734 - val_loss: 0.7286 - val_acc: 0.8297
Epoch 6/50
54000/54000 [==============================] - 1s - loss: 0.7208 - acc: 0.8192 - val_loss: 0.6151 - val_acc: 0.8647
Epoch 7/50
54000/54000 [==============================] - 1s - loss: 0.6218 - acc: 0.8495 - val_loss: 0.5358 - val_acc: 0.8845
Epoch 8/50
54000/54000 [==============================] - 1s - loss: 0.5518 - acc: 0.8664 - val_loss: 0.4613 - val_acc: 0.8970
Epoch 9/50
54000/54000 [==============================] - 1s - loss: 0.4973 - acc: 0.8773 - val_loss: 0.4200 - val_acc: 0.9015
Epoch 10/50
54000/54000 [==============================] - 1s - loss: 0.4537 - acc: 0.8867 - val_loss: 0.3881 - val_acc: 0.9102
Epoch 11/50
54000/54000 [==============================] - 1s - loss: 0.4244 - acc: 0.8926 - val_loss: 0.3479 - val_acc: 0.9160
Epoch 12/50
54000/54000 [==============================] - 1s - loss: 0.3913 - acc: 0.8990 - val_loss: 0.3307 - val_acc: 0.9163
Epoch 13/50
54000/54000 [==============================] - 1s - loss: 0.3747 - acc: 0.9008 - val_loss: 0.3198 - val_acc: 0.9200
Epoch 14/50
54000/54000 [==============================] - 1s - loss: 0.3559 - acc: 0.9042 - val_loss: 0.3076 - val_acc: 0.9178
Epoch 15/50
54000/54000 [==============================] - 1s - loss: 0.3416 - acc: 0.9092 - val_loss: 0.2982 - val_acc: 0.9228
Epoch 16/50
54000/54000 [==============================] - 1s - loss: 0.3351 - acc: 0.9092 - val_loss: 0.2883 - val_acc: 0.9222
Epoch 17/50
54000/54000 [==============================] - 1s - loss: 0.3254 - acc: 0.9100 - val_loss: 0.2785 - val_acc: 0.9257
Epoch 18/50
54000/54000 [==============================] - 1s - loss: 0.3134 - acc: 0.9136 - val_loss: 0.2789 - val_acc: 0.9270
Epoch 19/50
54000/54000 [==============================] - 1s - loss: 0.3087 - acc: 0.9141 - val_loss: 0.2524 - val_acc: 0.9295
Epoch 20/50
54000/54000 [==============================] - 1s - loss: 0.3067 - acc: 0.9135 - val_loss: 0.2622 - val_acc: 0.9247
Epoch 21/50
54000/54000 [==============================] - 1s - loss: 0.2891 - acc: 0.9188 - val_loss: 0.2459 - val_acc: 0.9338
Epoch 22/50
54000/54000 [==============================] - 1s - loss: 0.2813 - acc: 0.9193 - val_loss: 0.2438 - val_acc: 0.9332
Epoch 23/50
54000/54000 [==============================] - 1s - loss: 0.2826 - acc: 0.9204 - val_loss: 0.2444 - val_acc: 0.9335
Epoch 24/50
54000/54000 [==============================] - 1s - loss: 0.2713 - acc: 0.9228 - val_loss: 0.2208 - val_acc: 0.9400
Epoch 25/50
54000/54000 [==============================] - 1s - loss: 0.2680 - acc: 0.9232 - val_loss: 0.2231 - val_acc: 0.9395
Epoch 26/50
54000/54000 [==============================] - 1s - loss: 0.2630 - acc: 0.9247 - val_loss: 0.2298 - val_acc: 0.9347
Epoch 27/50
54000/54000 [==============================] - 1s - loss: 0.2635 - acc: 0.9242 - val_loss: 0.2280 - val_acc: 0.9345
Epoch 28/50
54000/54000 [==============================] - 1s - loss: 0.2527 - acc: 0.9286 - val_loss: 0.2105 - val_acc: 0.9417
Epoch 29/50
54000/54000 [==============================] - 1s - loss: 0.2460 - acc: 0.9296 - val_loss: 0.2020 - val_acc: 0.9465
Epoch 30/50
54000/54000 [==============================] - 1s - loss: 0.2466 - acc: 0.9278 - val_loss: 0.2037 - val_acc: 0.9425
Epoch 31/50
54000/54000 [==============================] - 1s - loss: 0.2373 - acc: 0.9307 - val_loss: 0.2058 - val_acc: 0.9425
Epoch 32/50
54000/54000 [==============================] - 1s - loss: 0.2483 - acc: 0.9278 - val_loss: 0.2157 - val_acc: 0.9347
Epoch 33/50
54000/54000 [==============================] - 1s - loss: 0.2503 - acc: 0.9291 - val_loss: 0.2073 - val_acc: 0.9452
Epoch 34/50
54000/54000 [==============================] - 1s - loss: 0.2394 - acc: 0.9305 - val_loss: 0.2097 - val_acc: 0.9413
Epoch 35/50
54000/54000 [==============================] - 1s - loss: 0.2341 - acc: 0.9310 - val_loss: 0.1993 - val_acc: 0.9423
Epoch 36/50
54000/54000 [==============================] - 1s - loss: 0.2316 - acc: 0.9323 - val_loss: 0.2063 - val_acc: 0.9412
Epoch 37/50
54000/54000 [==============================] - 1s - loss: 0.2300 - acc: 0.9330 - val_loss: 0.1971 - val_acc: 0.9432
Epoch 38/50
54000/54000 [==============================] - 1s - loss: 0.2259 - acc: 0.9338 - val_loss: 0.1969 - val_acc: 0.9440
Epoch 39/50
54000/54000 [==============================] - 1s - loss: 0.2211 - acc: 0.9356 - val_loss: 0.1933 - val_acc: 0.9442
Epoch 40/50
54000/54000 [==============================] - 1s - loss: 0.2233 - acc: 0.9345 - val_loss: 0.2032 - val_acc: 0.9432
Epoch 41/50
54000/54000 [==============================] - 1s - loss: 0.2238 - acc: 0.9331 - val_loss: 0.1878 - val_acc: 0.9468
Epoch 42/50
54000/54000 [==============================] - 1s - loss: 0.2141 - acc: 0.9383 - val_loss: 0.1871 - val_acc: 0.9463
Epoch 43/50
54000/54000 [==============================] - 1s - loss: 0.2135 - acc: 0.9380 - val_loss: 0.1871 - val_acc: 0.9475
Epoch 44/50
54000/54000 [==============================] - 1s - loss: 0.2135 - acc: 0.9384 - val_loss: 0.1826 - val_acc: 0.9488
Epoch 45/50
54000/54000 [==============================] - 1s - loss: 0.2073 - acc: 0.9402 - val_loss: 0.1847 - val_acc: 0.9468
Epoch 46/50
54000/54000 [==============================] - 1s - loss: 0.2120 - acc: 0.9371 - val_loss: 0.1765 - val_acc: 0.9498
Epoch 47/50
54000/54000 [==============================] - 1s - loss: 0.2059 - acc: 0.9408 - val_loss: 0.1786 - val_acc: 0.9502
Epoch 48/50
54000/54000 [==============================] - 1s - loss: 0.1995 - acc: 0.9419 - val_loss: 0.1793 - val_acc: 0.9483
Epoch 49/50
54000/54000 [==============================] - 1s - loss: 0.2044 - acc: 0.9414 - val_loss: 0.1781 - val_acc: 0.9450
Epoch 50/50
54000/54000 [==============================] - 1s - loss: 0.2047 - acc: 0.9403 - val_loss: 0.1987 - val_acc: 0.9403
First 3 predictions: [5 0 4]
Training accuracy: 93.31%
Test accuracy: 92.47%