import tensorflow as tf
from numpy.random import RandomState
batch_size = 8
# 兩個(gè)輸入節(jié)點(diǎn)
x = tf.placeholder(tf.float32, shape=(None, 2), name="x-input")
# 回歸問(wèn)題一般只有一個(gè)輸出節(jié)點(diǎn)
y_ = tf.placeholder(tf.float32, shape=(None, 1), name="y-input")
# 定義了一個(gè)單層的神經(jīng)網(wǎng)絡(luò)前向傳播的過(guò)程，這里就是簡(jiǎn)單加權(quán)和
w1 = tf.Variable(tf.random_normal([2, 1], stddev=1, seed=1))
y = tf.matmul(x, w1)
# 定義預(yù)測(cè)多了和預(yù)測(cè)少了的成本
loss_less = 10
loss_more = 1
#在windows下，下面用這個(gè)where替代，因?yàn)檎{(diào)用tf.select會(huì)報(bào)錯(cuò)
loss = tf.reduce_sum(tf.where(tf.greater(y, y_), (y - y_)*loss_more, (y_-y)*loss_less))
train_step = tf.train.AdamOptimizer(0.001).minimize(loss)
#通過(guò)隨機(jī)數(shù)生成一個(gè)模擬數(shù)據(jù)集
rdm = RandomState(1)
dataset_size = 128
X = rdm.rand(dataset_size, 2)
"""
設(shè)置回歸的正確值為兩個(gè)輸入的和加上一個(gè)隨機(jī)量，之所以要加上一個(gè)隨機(jī)量是
為了加入不可預(yù)測(cè)的噪音，否則不同損失函數(shù)的意義就不大了，因?yàn)椴煌瑩p失函數(shù)
都會(huì)在能完全預(yù)測(cè)正確的時(shí)候最低。一般來(lái)說(shuō)，噪音為一個(gè)均值為0的小量，所以
這里的噪音設(shè)置為-0.05， 0.05的隨機(jī)數(shù)。
"""
Y = [[x1 + x2 + rdm.rand()/10.0-0.05] for (x1, x2) in X]
with tf.Session() as sess:
 init = tf.global_variables_initializer()
 sess.run(init)
 steps = 5000
 for i in range(steps):
  start = (i * batch_size) % dataset_size
  end = min(start + batch_size, dataset_size)
  sess.run(train_step, feed_dict={x:X[start:end], y_:Y[start:end]})
 print(sess.run(w1))

[[ 1.01934695]
[ 1.04280889]