import tensorflow as tf
import numpy as np
import os
import matplotlib
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
def base_layer(input,in_size,out_size,activation_fun = None): # input个神经元 output神经元
w = tf.Variable(tf.random_normal([in_size,out_size])) # 生成一个大小为[in_size,out_size]的权重矩阵
b = tf.Variable(tf.zeros([1,out_size])+0.1) # 生成一个大小为[1,out_size]的偏差矩阵,元素值都为0.1
y = tf.matmul(input,w)+b # n行1列 乘以 1行10列 = n行10列 +1行10列 = n行10列
if activation_fun is None:
output = y
else:
output =activation_fun(y)
return output
if __name__=="__main__":
x_data = np.linspace(-1,1,10).astype('float32')[:,np.newaxis]
noise = np.random.normal(0,0.05,x_data.shape)
y_data = np.square(x_data)-0.5 + noise
# placeholder()函数是在神经网络构建graph的时候在模型中的占位,此时并没有把要输入的数据传入模型,它只会分配必要的内存。等建立session,在会话中,运行模型的时候通过feed_dict()函数向占位符喂入数据
xs = tf.placeholder(tf.float32,[None,1]) # N行 1列
ys = tf.placeholder(tf.float32,[None,1])
l1 = base_layer(xs,1,10,activation_fun=tf.nn.relu) #隐层输入xs,隐层有1个输入单元,10个输出单元,激活函数为relu.
result = base_layer(l1,10,1,activation_fun = None) #输出层输入l1,10个输入单元,1个输出单元,激活函数None(默认为线性)
loss = tf.reduce_mean(tf.reduce_sum(tf.square(ys-result),reduction_indices=[1]))
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
for i in range(201):
sess.run(train_step,feed_dict = {xs:x_data,ys:y_data})
if i % 50 ==0:
print(sess.run(loss,feed_dict = {xs:x_data,ys:y_data}))
sess.close()