TensorFlow筆記--求解二次函數最小值對應的值

最近學習在 TensorFlow，看了兩行代碼後覺得TensorFlow真是一個不錯的工具。然後，作爲工科男我就想着解決一些有趣的問題；解決什麼好呢？想來想去還是決定對我們熟悉的二次方程求解。我想大家對如何解二次方程早就成竹在胸，下面看一下TensorFlow如何做的吧。

問題：

我們先來看一下需要求解的方程：y = (x+1)^2；
然後給出我們的要求：求解在 y 最小時 x 的值。
那麼如何求解呢？OK，不賣關子啦，直接上代碼嘍！

代碼段：

# coding:utf-8
# 設函數 y = (x+1)^2, 令 x 初始值爲5
# 反向傳播就是求最優 x，即求最小y值對應的x值
import tensorflow as tf

# 定義待優化參數 x 初值爲 5
x = tf.Variable(tf.constant(5, dtype=tf.float32))
# y = (x+1)^2
y = tf.square(x+1)
# 定義反向傳播方法
train_step = tf.train.GradientDescentOptimizer(0.2).minimize(y)

# 生成會話，訓練40輪
with tf.Session() as sess:
    init_op = tf.global_variables_initializer()
    sess.run(init_op)
    for i in range(40):
        sess.run(train_step)
        x_val = sess.run(x)
        y_val = sess.run(y)
        print("After %s steps: x is %f,  y is %f."%(i, x_val, y_val))

運行結果如下：

After 0 steps: x is 2.600000, y is 12.959999.
After 1 steps: x is 1.160000, y is 4.665599.
After 2 steps: x is 0.296000, y is 1.679616.
After 3 steps: x is -0.222400, y is 0.604662.
After 4 steps: x is -0.533440, y is 0.217678.
After 5 steps: x is -0.720064, y is 0.078364.
After 6 steps: x is -0.832038, y is 0.028211.
After 7 steps: x is -0.899223, y is 0.010156.
After 8 steps: x is -0.939534, y is 0.003656.
After 9 steps: x is -0.963720, y is 0.001316.
After 10 steps: x is -0.978232, y is 0.000474.
After 11 steps: x is -0.986939, y is 0.000171.
After 12 steps: x is -0.992164, y is 0.000061.
After 13 steps: x is -0.995298, y is 0.000022.
After 14 steps: x is -0.997179, y is 0.000008.
After 15 steps: x is -0.998307, y is 0.000003.
After 16 steps: x is -0.998984, y is 0.000001.
After 17 steps: x is -0.999391, y is 0.000000.
After 18 steps: x is -0.999634, y is 0.000000.
After 19 steps: x is -0.999781, y is 0.000000.
After 20 steps: x is -0.999868, y is 0.000000.
After 21 steps: x is -0.999921, y is 0.000000.
After 22 steps: x is -0.999953, y is 0.000000.
After 23 steps: x is -0.999972, y is 0.000000.
After 24 steps: x is -0.999983, y is 0.000000.
After 25 steps: x is -0.999990, y is 0.000000.
After 26 steps: x is -0.999994, y is 0.000000.
After 27 steps: x is -0.999996, y is 0.000000.
After 28 steps: x is -0.999998, y is 0.000000.
After 29 steps: x is -0.999999, y is 0.000000.
After 30 steps: x is -0.999999, y is 0.000000.
After 31 steps: x is -1.000000, y is 0.000000.
After 32 steps: x is -1.000000, y is 0.000000.
After 33 steps: x is -1.000000, y is 0.000000.
After 34 steps: x is -1.000000, y is 0.000000.
After 35 steps: x is -1.000000, y is 0.000000.
After 36 steps: x is -1.000000, y is 0.000000.
After 37 steps: x is -1.000000, y is 0.000000.
After 38 steps: x is -1.000000, y is 0.000000.
After 39 steps: x is -1.000000, y is 0.000000.

就這樣利用 TensorFlow 刷了一題二次函數，是不是很開心呢？
或許看了這些你覺得TensorFlow被大材小用了，但是作爲TensorFlow的入門例子已經不簡單嘍。後面將會持續更新TensorFlow的使用，若你喜歡請添加關注哦！同時也歡迎您的留言，讓我們一起交流TensorFlow學習心得！