import tensorflow as tf
import os
os.environ['TF_LOG_MIN_LEVEL'] = '2'
class LinearRegression():
def __init__(self,data=None):
if data is None:
self.X,self.Y = self.gener_data()
else:
self.X,self.Y = data
self.weights = tf.Variable(initial_value=tf.random_normal(shape=[1,1],mean=0,stddev=0.1),dtype=tf.float32)
self.bias = tf.Variable(initial_value=tf.random_normal(shape=[1,1],mean=0,stddev=0.1),dtype=tf.float32)
with tf.get_default_graph().device("/gpu:0"):
self.y_ped = tf.matmul(self.X,self.weights)+self.bias
# 损失函数
self.loss = tf.reduce_mean(tf.square(self.Y-self.y_ped))
# 定义损失函数
self.optim = tf.train.GradientDescentOptimizer(learning_rate=0.001).minimize(self.loss)
self.init_data = tf.global_variables_initializer()
def fit(self,epochs=1000):
print("model training....")
with tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=True
)) as sess:
sess.run(self.init_data)
print("初始化变量：Weithts = %f, bias = %f, loss = %f" %
(sess.run(self.weights), sess.run(self.bias), sess.run(self.loss)))
for i in range(epochs):
#优化
sess.run(self.optim)
if (i + 1) % 1 == 0:
print("训练后第%d次后：Weithts = %f, bias = %f, loss = %f" %
(i + 1, sess.run(self.weights), sess.run(self.bias), sess.run(self.loss)))
def predict(self,x):
return tf.matmul(x,self.weights)+self.bias
def gener_data(self):
X = tf.random_normal(shape=[100,1],dtype=tf.float32)
noise = tf.random_normal(shape=[100,1],dtype=tf.float32)/1000.
# tf.case
Y = tf.matmul(X,[[2.0]])+0.5+noise
return X,Y
if __name__ == '__main__':
model = LinearRegression()
model.fit(epochs=10000)
```
