在這裡，我們訓練一個網路來擬合 y=x^2+1
程式碼如下：

# y = x^2 + 1

import tensorflow as tf
import numpy as np
import random

def get_batch(size=128):
    xs = []
    ys = []
    for i in range(size):
        x = random.random() * 2
        y = x * x + 1
        xs.append(x)
        ys.append(y)
    return np.array(xs), np.array(ys)




X = tf.placeholder(tf.float32, [None
 
,1], name='input')
Y = tf.placeholder(tf.float32, [None,1])
def my_dnn():
    x = tf.reshape(X, shape=[-1, 1])
    w1 = tf.Variable(tf.random_normal(shape=[1,256], mean=0.0,
                                      stddev=1))
    b1 = tf.Variable(tf.random_normal([256]))
    out1 = tf.nn.bias_add(tf.matmul(x,w1),b1)
    out1 = tf.nn.relu(out1)
    w2= tf.Variable(tf.random_normal(shape=[256
 
,256]))
    b2 = tf.Variable(tf.random_normal([256]))
    out2= tf.nn.bias_add(tf.matmul(out1, w2),b2)
    out2 = tf.nn.relu(out2)
    w3 = tf.Variable(tf.random_normal(shape=[256, 256]))
    b3 = tf.Variable(tf.random_normal([256]))
    out3 = tf.nn.bias_add(tf.matmul(out2, w3),b3)
    out3 = tf.nn.relu(out3)
    w4 = tf.Variable(tf.random_normal(shape=[256
 
, 1]))
    b4 = tf.Variable(tf.random_normal([1]))
    out4 = tf.nn.bias_add(tf.matmul(out3, w4), b4, name='output')


    return out4
def train():
    out = my_dnn()
    loss = tf.reduce_mean(tf.square(Y - out))
    optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)

    saver = tf.train.Saver()
    with tf.Session() as sess:
        sess.run(tf.initialize_all_variables())
        step = 0
        while True:
            batch_x, batch_y = get_batch(64)
            batch_x = batch_x.reshape([-1, 1])
            batch_y = batch_y.reshape([-1, 1])
            _, loss_ = sess.run([optimizer, loss], feed_dict={X:batch_x, Y:batch_y})
            print(loss_)
            if loss_ < 0.0001:
                saver.save(sess, "./1.model", global_step=step)
                break
            step += 1


# train()

def eval():
    out = my_dnn()
    saver = tf.train.Saver()
    with tf.Session() as sess:
        saver.restore(sess, tf.train.latest_checkpoint('.'))
        for i in range(100):
            x = random.random() * 2
            x = np.array([x]).reshape([-1,1])
            y = sess.run(out, feed_dict={X:x})
            print("x=%.5f 正確的y=%.5f 預測的 y=%.5f" % (x, x*x + 1, y))


if __name__ == '__main__':
    # 訓練
    # train()

    # 評估
    # eval()