Tensorflow的练习Demo

第一次练习：

import tensorflow as tf
import numpy as np

#creat data
x_data = np.random.rand(100).astype(np.float32)#产生[0,1]之间的100个随机数
y_data = np.multiply(x_data, x_data)*0.1 + x_data*0.2 + 0.3#y=0.1x^2 + 0.2x + 0.3

#creat tensorflow struchtre start
a = tf.Variable(tf.random_uniform([1], -1, 1))
b = tf.Variable(tf.random_uniform([1], -1, 1))
c = tf.Variable(tf.random_uniform([1], -1, 1))

y = a*np.multiply(x_data, x_data) + b*x_data + c
loss = tf.reduce_mean(tf.square(y-y_data))#误差
optimizer = tf.train.GradientDescentOptimizer(0.65)#选择了梯度下降的优化方法。0.65是学习效率，是0到1的数。越大收敛越快。

train = optimizer.minimize(loss)
init = tf.global_variables_initializer()

#creat tensorflow struchtre end
sess = tf.Session()
sess.run(init)#Very important

for step in range(1001):
    sess.run(train)
    if(step%20 == 0):
        print("step:%d a:%f b:%f c:%f"%(step, sess.run(a), sess.run(b), sess.run(c)))

step:0 a:-0.326912 b:0.021422 c:0.528646
step:20 a:0.010856 b:0.276253 c:0.288528
step:40 a:0.025957 b:0.279414 c:0.284537
step:60 a:0.030900 b:0.274777 c:0.285513
step:80 a:0.035218 b:0.270136 c:0.286428
step:100 a:0.039254 b:0.265769 c:0.287275
step:120 a:0.043037 b:0.261674 c:0.288067
step:140 a:0.046584 b:0.257833 c:0.288811
step:160 a:0.049911 b:0.254231 c:0.289507
step:180 a:0.053031 b:0.250854 c:0.290161
step:200 a:0.055956 b:0.247687 c:0.290774
step:220 a:0.058699 b:0.244717 c:0.291348
step:240 a:0.061271 b:0.241932 c:0.291887
step:260 a:0.063683 b:0.239320 c:0.292392
step:280 a:0.065944 b:0.236872 c:0.292866
step:300 a:0.068065 b:0.234575 c:0.293310
step:320 a:0.070054 b:0.232422 c:0.293727
step:340 a:0.071919 b:0.230403 c:0.294118
step:360 a:0.073668 b:0.228510 c:0.294484
step:380 a:0.075308 b:0.226734 c:0.294828
step:400 a:0.076846 b:0.225069 c:0.295150
step:420 a:0.078288 b:0.223508 c:0.295452
step:440 a:0.079640 b:0.222044 c:0.295735
step:460 a:0.080908 b:0.220671 c:0.296001
step:480 a:0.082097 b:0.219384 c:0.296250
step:500 a:0.083212 b:0.218176 c:0.296483
step:520 a:0.084257 b:0.217044 c:0.296702
step:540 a:0.085238 b:0.215983 c:0.296908
step:560 a:0.086157 b:0.214988 c:0.297100
step:580 a:0.087019 b:0.214054 c:0.297281
step:600 a:0.087828 b:0.213179 c:0.297450
step:620 a:0.088586 b:0.212358 c:0.297609
step:640 a:0.089297 b:0.211589 c:0.297758
step:660 a:0.089963 b:0.210867 c:0.297897
step:680 a:0.090588 b:0.210190 c:0.298028
step:700 a:0.091174 b:0.209555 c:0.298151
step:720 a:0.091724 b:0.208960 c:0.298266
step:740 a:0.092239 b:0.208402 c:0.298374
step:760 a:0.092723 b:0.207879 c:0.298476
step:780 a:0.093176 b:0.207388 c:0.298571
step:800 a:0.093601 b:0.206928 c:0.298660
step:820 a:0.093999 b:0.206497 c:0.298743
step:840 a:0.094373 b:0.206092 c:0.298821
step:860 a:0.094724 b:0.205713 c:0.298895
step:880 a:0.095052 b:0.205357 c:0.298964
step:900 a:0.095360 b:0.205023 c:0.299028
step:920 a:0.095649 b:0.204711 c:0.299089
step:940 a:0.095920 b:0.204417 c:0.299145
step:960 a:0.096174 b:0.204142 c:0.299199
step:980 a:0.096413 b:0.203884 c:0.299249
step:1000 a:0.096636 b:0.203642 c:0.299295

总体上a、b、c越来越接近真实傎。
查了一下tf.random_uniform的用法：

tf.random_uniform((4, 4), minval=-0.5,maxval=0.5,dtype=tf.float32)))