[机器学习] keras：MNIST手写数字体识别（DeepLearning 的 HelloWord程序）

深度学习界的Hello Word程序：MNIST手写数字体识别
learn from（仍然是李宏毅老师《机器学习》课程）：http://speech.ee.ntu.edu.tw/~tlkagk/courses_ML17_2.html

import numpy as np 
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation
from keras.layers import Conv2D, MaxPooling2D, Flatten
from keras.optimizers import SGD, Adam
from keras.utils import np_utils
from keras.datasets import mnist

#定义数据加载、预处理函数
def load_data():
    (x_train, y_train), (x_test, y_test) = mnist.load_data()
    number = 10000
    x_train = x_train[0:number]
    y_train = y_train[0:number]
    x_train = x_train.reshape(number, 28*28)
    x_test = x_test.reshape(x_test.shape[0], 28*28)
    x_train = x_train.astype('float32')
    x_test = x_test.astype('float32')
#keras.utils.to_categorical(y, num_classes=None, dtype='float32')
#自动处理成one-hot的numpy数组。y为标签数据，num_classes是一共有几类）
    y_train = np_utils.to_categorical(y_train, 10)
    y_test = np_utils.to_categorical(y_test, 10)

    x_train = x_train/255   #图像数据预处理 标准化
    x_test = x_test/255
    return (x_train, y_train), (x_test, y_test)


(x_train, y_train), (x_test, y_test) = load_data()

#搭建网络
model = Sequential()
model.add(Dense(input_dim=28*28, units=689, activation='relu'))
model.add(Dense(units=689, activation='relu'))
model.add(Dense(units=10, activation='softmax'))
model.compile(loss='categorical_crossentropy',optimizer=SGD(lr=0.1),metrics=['accuracy'])
#训练走起！
model.fit(x_train, y_train, batch_size=50, epochs=20)
#看train出模型在test data上的表现如何
result = model.evaluate(x_test,y_test)
print('Test Acc:',result[1])

训练量很小，我们就不切GPU训了。效果：

Using TensorFlow backend.
Epoch 1/20
10000/10000 [==============================] - 2s 166us/step - loss: 0.5926 - acc: 0.8301
Epoch 2/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.2686 - acc: 0.9234
Epoch 3/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.2015 - acc: 0.9407
Epoch 4/20
10000/10000 [==============================] - 1s 106us/step - loss: 0.1546 - acc: 0.9568
Epoch 5/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.1238 - acc: 0.9645
Epoch 6/20
10000/10000 [==============================] - 1s 106us/step - loss: 0.0977 - acc: 0.9744
Epoch 7/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0808 - acc: 0.9790
Epoch 8/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0638 - acc: 0.9838
Epoch 9/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0514 - acc: 0.9875
Epoch 10/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0424 - acc: 0.9905
Epoch 11/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0331 - acc: 0.9936
Epoch 12/20
10000/10000 [==============================] - 1s 108us/step - loss: 0.0267 - acc: 0.9960
Epoch 13/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0209 - acc: 0.9972
Epoch 14/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0177 - acc: 0.9977
Epoch 15/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0144 - acc: 0.9986
Epoch 16/20
10000/10000 [==============================] - 1s 106us/step - loss: 0.0115 - acc: 0.9993
Epoch 17/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0102 - acc: 0.9994
Epoch 18/20
10000/10000 [==============================] - 1s 106us/step - loss: 0.0084 - acc: 0.9997
Epoch 19/20
10000/10000 [==============================] - 1s 106us/step - loss: 0.0071 - acc: 0.9998
Epoch 20/20
10000/10000 [==============================] - 1s 107us/step - loss: 0.0064 - acc: 0.9999
10000/10000 [==============================] - 0s 38us/step
Test Acc: 0.9573