keras 识别Mnist

# -*- coding: utf-8 -*-
# Larger CNN for the MNIST Dataset  
# 2.Negative dimension size caused by subtracting 5 from 1 for 'conv2d_4/convolution' (op: 'Conv2D') with input shapes  
# 3.UserWarning: Update your `Conv2D` call to the Keras 2 API: http://blog.csdn.net/johinieli/article/details/69222956  
# 4.Error when checking input: expected conv2d_1_input to have shape (None, 28, 28, 1) but got array with shape (60000, 1, 28, 28)  
 
# talk to wumi,you good .  
 
# python 3.5.4  
# keras.__version__  : 2.0.6  
# thensorflow 1.2.1  
# theano 0.10.0beta1  
 
# 不错的blog  http://blog.csdn.net/shizhengxin123/article/details/72383728  
 
import numpy  
from keras.datasets import mnist  
from keras.models import Sequential  
from keras.layers import Dense  
from keras.layers import Dropout  
from keras.layers import Flatten  
from keras.layers.convolutional import Conv2D  
from keras.layers.convolutional import MaxPooling2D  
from keras.utils import np_utils  
import matplotlib.pyplot as plt  
from keras.constraints import maxnorm  
from keras.optimizers import SGD  
# fix random seed for reproducibility  
seed = 7  
numpy.random.seed(seed)  
# load data  
(X_train, y_train), (X_test, y_test) = mnist.load_data()  
# reshape to be [samples][pixels][width][height]  
X_train = X_train.reshape(X_train.shape[0], 28, 28,1).astype('float32')  
X_test = X_test.reshape(X_test.shape[0],28, 28,1).astype('float32')  
#X_train = X_train.reshape(X_train.shape[0], 1, 28, 28).astype('float32')  
#X_test = X_test.reshape(X_test.shape[0], 1, 28, 28).astype('float32')    <---4   
# normalize inputs from 0-255 to 0-1  
X_train = X_train / 255  
X_test = X_test / 255  
# one hot encode outputs  
y_train = np_utils.to_categorical(y_train)  
y_test = np_utils.to_categorical(y_test)  
num_classes = y_test.shape[1]  
###raw  
# define the larger model  
def larger_model():  
    # create model  
    model = Sequential()  
    model.add(Conv2D(30, (5, 5), padding='valid', input_shape=(28, 28,1), activation='relu'))  
    #model.add(Conv2D(30, (5, 5), padding='valid', input_shape=(28, 28,1), activation='relu'))   <----3,2  
    model.add(MaxPooling2D(pool_size=(2, 2)))  
    model.add(Dropout(0.4))  
    model.add(Conv2D(15, (3, 3), activation='relu'))  
    model.add(MaxPooling2D(pool_size=(2, 2)))  
    model.add(Dropout(0.4))  
    model.add(Flatten())  
    model.add(Dense(128, activation='relu'))  
    model.add(Dropout(0.4))  
    model.add(Dense(50, activation='relu'))  
    model.add(Dropout(0.4))  
    model.add(Dense(num_classes, activation='softmax'))  
    # Compile model  
    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])  
    return model  
 
# build the model  
model = larger_model()  
# Fit the model  
model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=10, batch_size=200, verbose=2)   # epochs 200 too bigger  
#model.fit(X_train, y_train, validation_data=(X_test, y_test), nb_epoch=200, batch_size=200, verbose=2)  
# Final evaluation of the model  
scores = model.evaluate(X_test, y_test, verbose=0)  
print("Large CNN Error: %.2f%%" % (100-scores[1]*100))  
scores1 = model.evaluate(X_test[0], y_test[0], verbose=0)  
print(scores1)