Tensorflow函数拟合sin

代码： 

import  tensorflow as tf
import matplotlib.pyplot as pyplot

fileQueue = tf.train.string_input_producer(["e:\\simdata.tf"])

tfReader = tf.TFRecordReader()
_, item = tfReader.read(fileQueue)
feature = tf.parse_single_example(item, features={"x":tf.FixedLenFeature([], tf.float32),
                                        "y": tf.FixedLenFeature([], tf.float32)}) 

a = tf.placeholder(shape=[None,1],name="input", dtype=tf.float32)
b = tf.placeholder(shape=[None,1],name="input", dtype=tf.float32)

Layer = tf.layers.dense(a, 32, tf.nn.tanh,kernel_initializer=tf.random_normal_initializer(stddev=0.1)) 
Layer1= tf.layers.dense(Layer, 32, tf.nn.tanh ,kernel_initializer=tf.random_normal_initializer(stddev=0.1) ) 
Layer2= tf.layers.dense(Layer1, 32, tf.nn.tanh ,kernel_initializer=tf.random_normal_initializer(stddev=0.1) )
Layer3= tf.layers.dense(Layer2, 1, tf.nn.tanh ,kernel_initializer=tf.random_normal_initializer(stddev=0.1) ) 
loss =  tf.reduce_mean(tf.square( Layer3 - b))

global_step = tf.Variable(0)
learning_rate = tf.train.exponential_decay(0.01, global_step, 1000, 0.9, staircase=True)

train = tf.train.AdamOptimizer(learning_rate).minimize(loss,global_step=global_step)

sw = tf.summary.FileWriter("e:/log")
tf.summary.scalar("loss", loss)
summall = tf.summary.merge_all()

saver = tf.train.Saver(max_to_keep=2);

with tf.Session() as sess:
    tf.global_variables_initializer().run()
    saver.restore(sess,"e:/save/save1.data-54100")
    tf.train.start_queue_runners(sess=sess)


    lossAry = [];
    losstemp = 0;
    runary = []
    runnumber = 0

    checkX = []
    checkY=[]

    for i in range(-314, 314,1):
        temp = (sess.run(feature));
        checkX.append([temp["x"]])
        checkY.append([temp["y"]])

    for t in range(1000):
        for j in range(100):
            tempA = []
            tempB = []
            for i in range(100):
                temp = (sess.run(feature));
                tempA.append([temp["x"]])
                tempB.append([temp["y"]])

            _ ,losstemp,summallTemp= sess.run([train,loss,summall], feed_dict={ a:tempA,
                                          b:tempB})

            sw.add_summary(summallTemp, j)

        runnumber = runnumber + 100

        runary.append(runnumber)
        temp = sess.run(loss, feed_dict={a:checkX,  b:checkY})
        lossAry.append(temp)

        pyplot.cla()
        pyplot.plot(runary,lossAry  )
        pyplot.show()

        saver.save(sess, "e:/save/save1.data",global_step=runnumber)
sw.close();

损失函数下降曲线：

效果