2.6 a easy machine learning ML

import tensorflow as tf
sess = tf.Session()
import numpy as np

#  production a   x

x = np.random.normal(1,0.1,100)
x
# make num 100  promix 1.0 of data

array([0.89772354, 1.01282827, 0.8221541 , 1.03343903, 1.09652531,
       1.08504872, 0.99351567, 0.94295761, 1.07872768, 1.09542048,
       0.86140521, 0.96496625, 1.09899229, 1.12631152, 0.85821607,
       0.98690985, 0.79504865, 0.98037352, 0.9416307 , 0.93330317,
       0.93686878, 0.74042065, 0.82757151, 0.90489619, 1.13785211,
       1.00790868, 0.84631589, 0.90302969, 0.96682828, 0.93569319,
       1.07605247, 1.00843382, 1.13323741, 1.0999531 , 0.97425667,
       1.05045277, 0.85205309, 1.22594526, 0.88120865, 0.85577489,
       1.0212521 , 1.01028048, 0.82811159, 1.25132735, 1.04923821,
       0.9660119 , 0.99091055, 0.9633072 , 0.94239817, 0.77503792,
       0.95303095, 1.15420376, 0.9489185 , 0.92237038, 0.76171667,
       0.91276112, 1.02078566, 1.17075822, 0.88780699, 1.11376264,
       0.85125444, 1.10461272, 0.85621748, 1.10177113, 1.07613838,
       1.02658167, 0.94826485, 1.04867475, 0.99410329, 1.01212242,
       1.09523441, 1.00079448, 1.08044779, 0.97345393, 1.2627164 ,
       0.99894025, 0.96131037, 0.86729808, 1.01904677, 1.09313799,
       0.94112591, 1.07627833, 1.08323275, 0.85524709, 0.84006658,
       1.00688393, 0.98967267, 1.12536786, 0.93636957, 1.03461993,
       0.89944818, 1.0997615 , 0.83533433, 1.05027472, 1.11580271,
       0.96905061, 0.97344286, 0.93910161, 0.92957192, 0.95551175])

y = np.repeat(10,100)
y
# make  100`s  10

array([10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
       10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
       10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
       10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
       10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
       10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10])

# x- y  relation  is  10
A = tf.Variable(tf.random_normal(shape=[1]))
sess.run(tf.initialize_all_variables())
print(sess.run(A))

[-0.33922198]

#sess.run(tf.initialize_all_variables())  use once change once variables
print(sess.run(A))  

[-0.33922198]

x_input = tf.placeholder(shape=[1],dtype=tf.float32)
y_pred = tf.multiply(A,x_input)
#print(sess.run(y_pred),feed_dict={x_input:x[0]}) 

y_output = tf.placeholder(shape=[1],dtype=tf.float32)
loss = tf.square(y_output-y_pred)

train_step = tf.train.GradientDescentOptimizer(learning_rate=0.02).minimize(loss)
sess.run(train_step,feed_dict={x_input:[1],y_output:[10]})

print(sess.run(train_step,feed_dict={x_input:[1],y_output:[10]}))

None

print(sess.run(A))
print(sess.run(y_pred,feed_dict={x_input:[1],y_output:[10]}))
print(sess.run(loss,feed_dict={x_input:[1],y_output:[10]}))

[3.126148]
[3.126148]
[47.24984]

凡是带tf 的 都需要再sess里运行
feed 数据后， 开始计算
train 一次改变一次， 变量；
那边变量的定义，以及 placeholder 的数据 输入 应该是在图之外的 。
打开tensorboard 看一下