深度学习之Python 脚本训练keras mnist 数字识别模型

本脚本是训练keras 的mnist 数字识别程序 ，以前发过了 ，今天把 预测实现了，

# 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

# good blog
# http://blog.csdn.net/shizhengxin123/article/details/72383728
# http://www.360doc.com/content/17/0415/12/1489589_645772879.shtml

# recommand another framework  http://tflearn.org/examples/

import numpy
import keras
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
from keras.preprocessing import image
import skimage.io



# fix random seed for reproducibility
seed = 7
numpy.random.seed(seed)
# load data
(X_train, y_train), (X_test, y_test) = mnist.load_data()

plt.subplot(221)

plt.imshow(X_train[0], cmap=plt.get_cmap('gray'))

plt.show()
# 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(1, 28, 28, 1).astype('float32') ValueError: cannot reshape array of size 47040000 into shape (1,28,28,1)
#X_test = X_test.reshape(1, 28, 28, 1).astype('float32')  ValueError: cannot reshape array of size 47040000 into shape (1,28,28,1)
# 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
    # optimizer  优化器
    # loss 损失函数
    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
    return model


# build the model
model = larger_model()
# Fit the model
# fit函数返回一个History的对象，其History.history属性记录了损失函数和其他指标的数值随epoch变化的情况，如果有验证集的话，也包含了验证集的这些指标变化情况
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))

# save the model
model.save('D:\\works\\jetBrians\\PycharmProjects\\tryPicture\\my_model.h5')  # creates a HDF5 file 'my_model.h5'
del model

# reload the modle
# returns a compiled model
# identical to the previous one
# modelTrained = Sequential()
# model = modelTrained.load_model('D:\\works\\jetBrians\\PycharmProjects\\tryPicture\\my_model.h5')

# https://gist.github.com/ageitgey/a40dded08e82e59724c70da23786bbf0

# write a number in a picture
# predict numbers

#image_path = './lena.jpg'
# method 1
# load pic
#img = image.load_img(image_path, target_size=(28, 28))
# handle pic
#x = image.img_to_array(img)
#x = numpy.expand_dims(x, axis=0)
#x = preprocess_input(x)

# method2
#img2 = skimage.io.imread(image_path, as_grey=True)
#skimage.io.imshow(img2)
#plt.show()
#img2 = numpy.reshape(img2, (1, 28, 28, 1)).astype('float32')
# 对数字进行预测
#https://baijiahao.baidu.com/s?id=1574962680356106&wfr=spider&for=pc
#predict = model.predict(img2, verbose=0)
#result = model.prediect_classes(img2, verbose=0)
#print(predict[0])
#print(result[0])

#some warning tips  The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
#have no idea what's the meaning


原来数据样式 




=================训练log

D:\applications\Anaconda3\python.exe D:/works/jetBrians/PycharmProjects/tryPicture/trainModel/TrainModel.py
Using TensorFlow backend.
Train on 60000 samples, validate on 10000 samples
Epoch 1/10
62s - loss: 0.8830 - acc: 0.7027 - val_loss: 0.1566 - val_acc: 0.9545
Epoch 2/10
56s - loss: 0.3130 - acc: 0.9078 - val_loss: 0.0955 - val_acc: 0.9712
Epoch 3/10
61s - loss: 0.2342 - acc: 0.9340 - val_loss: 0.0737 - val_acc: 0.9763
Epoch 4/10
58s - loss: 0.1924 - acc: 0.9458 - val_loss: 0.0643 - val_acc: 0.9802
Epoch 5/10
60s - loss: 0.1678 - acc: 0.9534 - val_loss: 0.0541 - val_acc: 0.9848
Epoch 6/10
53s - loss: 0.1541 - acc: 0.9578 - val_loss: 0.0468 - val_acc: 0.9849
Epoch 7/10
53s - loss: 0.1396 - acc: 0.9617 - val_loss: 0.0464 - val_acc: 0.9852
Epoch 8/10
55s - loss: 0.1303 - acc: 0.9647 - val_loss: 0.0422 - val_acc: 0.9871
Epoch 9/10
52s - loss: 0.1276 - acc: 0.9656 - val_loss: 0.0398 - val_acc: 0.9871
Epoch 10/10
53s - loss: 0.1156 - acc: 0.9680 - val_loss: 0.0370 - val_acc: 0.9876
Large CNN Error: 1.24%

Process finished with exit code 0