#总结神经网络框架 #1,搭建网络设计结构(前向传播) 文件forward.py def forward(x,regularizer): # 输入x 和正则化权重 w= b= y= return y def get_weight(shape,regularizer): #w的shape 和w的权重 w=tf.Variable(tf.random_normal(shape),dtype=tf.float32) # 给w赋初值 #给 每一个w的正则化损失加到总损失中 tf.add_to_collection('losses',tf.contrib.layers.l2_regularizer(regularizer)(w)) return w def get_bias(shape): #某一层b的个数 b=tf.Variable(tf.constant(0.01,shape=shape)) #赋初值 return b #反向传播就是训练模型,优化参数 文件backward.py def backward(): x=tf.placeholder(tf.float32,shape=(None,2)) #给输入占位 y_=tf.placeholder(tf.float32,shape=(None,1)) y=forward.forward(x,refularizer) global_step=tf.Variable(0,trianable=False) #轮数计数器 loss= #定义loss # loss 可以是 自定义或者交叉熵 # y与y_的差距{loss_mse}=tf.reduce_mean(tf.square(y-y_)) # 也可以是: # ce=tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y,labels=tf.argmax(y_,1)) # y与y_的差距{cem}=tf.reduce_mean(ce) # 加入正则化之后,需要 # loss=y与y_的差距+tf.add_n(tf.get_collection('losses')) train_step=tf.train.GradientDescentOptimizer(learning_rate).minimize(loss,global_step=global_step) # 如果使用ema 增加以下,动态计算学习率 learning_rate=tf.train.exponential_decay(LEARNING_RATE_BASE, global_step, LEARNING_RATE_STEP(即数据库样本总数/batch_size), LEARNING_RATE_DECAY, staircase=True) ema=tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY,global_step) ema_op=ema.apply(tf.trainable_variables()) with tf.control_dependencies([train_step,ema_op]): train_op=tf.no_op(name='train') with tf.Session() as sess: init_op=tf.global_variables_initializer() sess.run(init_op) for i in range(steps): sess.run(train_step,feed_dict={x: ,y_= }) if i % 轮数 ==0: print('') if '__name__'=='__main__': #判断是否运行的是主文件 backward()
使用正则化提高泛化性, 使用指数衰减学习率加快优化效率
使用三个模块
1 生成数据集 generateds.py
2 前向传播 forward.py
3 反向传播 backward.py
# 生成数据集 generateds.py import numpy as np import matploblib.pyplot as plt seed=2 def genetateds(): rdm=np.random.RandomState(seed) X=rdm.randn(300,2) Y_=[int(x0*x0+x1*x1<2) for (x0,x1) in X] Y_c=[['red' if y else 'blue'] for y in Y_] #1则红色,0则蓝色 X=np.vstack(X).reshape(-1,2) #整理为n行2列,按行的顺序来 Y_=np.vstack(Y_).reshape(-1,1)# 整理为n行1列 return X,Y_,Y_c
# 前向传播 import tensorflow as tf def get_weight(shape,regularizer): #w的shape 和w的权重 w=tf.Variable(tf.random_normal(shape),dtype=tf.float32) tf.add_to_collection('losses',tf.contrib.layers.l2_regularizer(regularizer)(w)) return w def get_bias(shape): #b的长度 b=tf.Variable(tf.constant(0.01,shape=shape)) return b def forward(x,regularizer): # 输入x 和正则化权重 w1=get_weight([2,11],regularizer) b1=get_bias([11]) y1=tf.nn.relu(tf.matmul(x,w1)+b1) #relu 激活函数 w2=get_weight([11,1],regularizer) b2=get_bias([1]) y=tf.matmul(y1,w2)+b2 #输出层不过激活函数 return y
# 反向传播 import tensorflow as tf import numpy as np import matplotlib.pyplot as plt import genarateds import forward steps=40000 batch_size=30 LEARNING_RATE_BASE=0.001 LEARNING_RATE_DECAY=0.999 regularizer=0.01 def backward(): x=tf.placeholder(tf.float32,shape=(None,2)) #给输入占位 y_=tf.placeholder(tf.float32,shape=(None,1)) X,Y_,Y_c=genarateds.generateds() y=forward.forward(x,refularizer) global_step=tf.Variable(0,trianable=False) #轮数计数器 learning_rate=tf.train.exponential_decay(LEARNING_RATE_BASE, global_step, 300/batch_size, LEARNING_RATE_DECAY, staircase=True) #定义损失函数 loss_mse=tf.reduce_mean(tf.square(y-y_)) loss_total=loss_mse+tf.add_n(tf.get_collection('losses')) #定义反向传播方法 包含正则化 train_step=tf.train.AdamOptimizer(learning_rate).minimize(loss_total) with tf.Session() as sess: init_op=tf.global_variables_initializer() sess.run(init_op) for i in range(steps): start=(i*batch_size)%300 end=start+batch_size sess.run(train_step,feed_dict={x:X[start:end],y_:Y_[start:end]}) if i%10000==0: loss_v=sess.run(loss_total,feed_dict={x:X,y_:Y_}) print('after %d steps,loss is:%f'%(i,loss_v)) xx,yy=np.mgrid[-3:3:0.01,-3:3:0.01] grid=np.c_[xx.ravel(),yy.ravel()] probs=sess.run(y,feed_dict={x:grid}) probs=probs.reshape(xx.shape) #调整成xx的样子 plt.scatter(X[:,0],X[:,1],c=np.squeeze(Y_c)) plt.contour(xx,yy,probs,levels=[.5]) #给probs=0.5的值上色 (显示分界线) plt.show() if '__name__'=='__main__': #判断是否运行的是主文件 backward()
最后运行 backward.py 即可!!