keras theano 小栗子

数据集：minst
神经网络：三层模型
优化算法：SGD

# coding:utf-8from keras.optimizers import SGDfrom mykeras import minst_pklminst = minst_pkl()train_x, train_y, valid_x, valid_y, test_x, test_y = minst.getIMGSets()from keras.models import Sequentialfrom keras.layers.core import Dense, Activation# modelmodel = Sequential()model.add(Dense(386, input_dim=784, init='uniform'))model.add(Activation('sigmoid'))model.add(Dense(10, activation='softmax'))# compilesgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)model.compile(loss='categorical_crossentropy', optimizer=sgd)# trainmodel.fit(train_x, train_y, nb_epoch=10, batch_size=200)# evaluate# score = model.evaluate(test_x, test_y, batch_size=200)# print "loss=", score# predictprint numpy.argmax(model.predict(test_x[0:5, :]), axis=1)print numpy.argmax(test_y[0:5, :], axis=1)

import cPickleimport gzipimport numpy as npclass minst_pkl(object):    def loadIMGData(self, dataset):        f = gzip.open(dataset, 'rb')        train_set, valid_set, test_set = cPickle.load(f)        f.close()        return [train_set, valid_set, test_set]    def getIMGSets(self):        print "load dataset"        sets = self.loadIMGData("data/mnist.pkl.gz")        train_x, train_y = sets[0]        valid_x, valid_y = sets[1]        test_x, test_y = sets[2]        print "train image,label shape:", train_x.shape, train_y.shape        print "valid image,label shape:", valid_x.shape, valid_y.shape        print "test  image,label shape:", test_x.shape, test_y.shape        print "load dataset end"        return [train_x, self.transform(train_y), valid_x, self.transform(valid_y), test_x, self.transform(test_y)]    def transform(self, data):        m = np.shape(data)[0]        convert = np.zeros((m, 10))        for i in np.arange(m):            convert[i, data[i]] = 1        return convertif __name__ == "__main__":    minst = minst_pkl()    train_x, train_y, valid_x, valid_y, test_x, test_y = minst.getIMGSets()

实验结果：

Epoch 1/1050000/50000 [==============================] - 2s - loss: 0.5389     Epoch 2/1050000/50000 [==============================] - 2s - loss: 0.3043     Epoch 3/1050000/50000 [==============================] - 2s - loss: 0.2689     Epoch 4/1050000/50000 [==============================] - 2s - loss: 0.2357     Epoch 5/1050000/50000 [==============================] - 2s - loss: 0.2069     Epoch 6/1050000/50000 [==============================] - 2s - loss: 0.1814     Epoch 7/1050000/50000 [==============================] - 2s - loss: 0.1619     Epoch 8/1050000/50000 [==============================] - 2s - loss: 0.1457     Epoch 9/1050000/50000 [==============================] - 2s - loss: 0.1315     Epoch 10/1050000/50000 [==============================] - 2s - loss: 0.1195     [7 2 1 0 4][7 2 1 0 4]

Using Theano backend.
Using gpu device 0: GeForce 820M (CNMeM is disabled, cuDNN not available)
gpu：29.75
cpu：99.99