from keras.callbacks import TensorBoard
from keras.models import Sequential
from keras.optimizers import SGD, Adam
from keras.layers import Dense, Flatten, Dropout
from keras.layers.convolutional import Conv2D, MaxPooling2D
from keras.models import load_model
import keras
import numpy as np
from keras.applications.imagenet_utils import preprocess_input
from keras import backend as K
from keras.datasets import cifar10
from tensorflow.examples.tutorials.mnist import input_data
K.clear_session()
mnist = input_data.read_data_sets("MNIST_DATA", one_hot=True)
class AlexModel:
#初始化参数
def __init__(self, epochs, batch_size):
"""
:param epochs: 训练集迭代的轮数
:param batch_size: 每次训练的样本的个数
"""
self.epochs = epochs
self.batch_size = batch_size
# 存储训练过程中的精度和误差
self.train_accuracy_and_loss = None
# 创建模型
def build_model(self):
"""
创建模型, 基于alexnet
:return:
"""
model = Sequential()
#第一层卷积网络，使用96个卷积核，大小为11x11步长为4， 要求输入 1个通道，激活函数使用relu
model.add(Conv2D(96, (11, 11), strides=(4, 4), input_shape=(28, 28, 1), padding='valid', activation='relu',
kernel_initializer='uniform'))
# 池化层
model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='valid'))
# 第二层加边使用256个5x5的卷积核，加边，激活函数为relu
model.add(Conv2D(256, (5, 5), strides=(1, 1), padding='same', activation='relu', kernel_initializer='uniform'))
#使用池化层，步长为2
model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same'))
# 第三层卷积，大小为3x3的卷积核使用384个
model.add(Conv2D(384, (3, 3), strides=(1, 1), padding='same', activation='relu', kernel_initializer='uniform'))
# 第四层卷积,同第三层
model.add(Conv2D(384, (3, 3), strides=(1, 1), padding='same', activation='relu', kernel_initializer='uniform'))
# 第五层卷积使用的卷积核为256个，其他同上
model.add(Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu', kernel_initializer='uniform'))
model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same'))
# 将卷积展开为神经元
model.add(Flatten())
# 第1层隐藏全连接层使用4096个神经元
model.add(Dense(4096, activation='relu'))
# dropout正则化
model.add(Dropout(0.5))
# 第2层隐藏使用4096个神经元
model.add(Dense(4096, activation='relu'))
model.add(Dropout(0.5))
# 输出层输出类别个数
model.add(Dense(10, activation='softmax'))
# 选用adam优化器，学习率为0.0003
adam = Adam(lr=0.0003, decay=1e-6)
# 编译模型
model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy'])
return model
# 保存模型
def save_model_after_train(self):
model = self.build_model()
x_train, y_train = mnist.train.images, mnist.train.labels
x_train = x_train.reshape(55000, 28, 28, 1)
self.train_accuracy_and_loss = model.fit(x_train, y_train, batch_size=self.batch_size, epochs=self.epochs)
model.save("model.h5")
# 加载模型
def load_model(self):
return load_model("model.h5")
# 训练模型
def train(self, mnist):
modle = self.build_model()
x_train, y_train = mnist.train.images, mnist.train.labels
x_train = x_train.reshape(55000, 28, 28, 1)
# {'acc': [], 'loss': []}
self.train_accuracy_and_loss = modle.fit(x_train, y_train, batch_size=self.batch_size,
epochs=self.epochs,
callbacks=[TensorBoard(log_dir='mytensorboard/3')])
# 获取训练过程中的损失（每个epoch）
def get_train_loss(self):
return self.train_accuracy_and_loss.history["loss"]
# 获取训练过程中的精度(每个epoch)
def get_train_accuracy(self):
return self.train_accuracy_and_loss.history["acc"]
# 测试集的精度和误差
def test_accuracy_and_loss(self):
""""将训练好的模型直接拿过来用"
:return: 返回精度和损失
"""
model = self.load_model()
x_test, y_test= mnist.test.images, mnist.test.labels
x_test = x_test.reshape(10000, 28, 28, 1)
score = model.evaluate(x_test, y_test, batch_size=32)
return score[1], score[0]
model = AlexModel(epochs=2, batch_size=256)
model.train(mnist)
loss = model.get_train_loss()
acc = model.get_train_accuracy()
print(acc)
