我的回归网络不是学习

Question

我在具有张量流的最简单的回归网络中测试了我的数据，并且我没有得到有效的学习。这是我的代码：

import tensorflow as tf

with tf.variable_scope("Inputs"):
    tf_x = tf.placeholder(tf.float32, xs.shape, name="x")
    tf_y = tf.placeholder(tf.float32, ys.shape, name="y")

with tf.variable_scope("Net"):
    l1 = tf.layers.dense(tf_x, 3, tf.nn.relu)
    weights = l1.graph.get_tensor_by_name(os.path.split(l1.name)[0] + "/kernel:0")
    output = tf.layers.dense(l1, 1)
    tf.summary.histogram("h_out", l1)
    tf.summary.histogram("pred", output)
    tf.summary.histogram("weights", weights)

loss = tf.losses.mean_squared_error(tf_y, output, scope="loss")
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.05)
train_op = optimizer.minimize(loss)
tf.summary.scalar("loss", loss)
sess = tf.Session()
sess.run(tf.global_variables_initializer())

writer = tf.summary.FileWriter("./logs", sess.graph)
merge_op = tf.summary.merge_all()

for step in range(1000):
    _, l, pred, result = sess.run([train_op, loss, output, merge_op], {tf_x:xs, tf_y:ys})
    writer.add_summary(result, step)
    if step % 100 == 0:
        print (l)

我的数据是：

x:   
     2.120
     1.860
     2.310
     2.060
     2.520
     1.770
     1.450
     1.420
     2.250
     1.930
     2.550
     2.050
     2.250
     2.570
     1.790
     2.380
     2.570
     1.850
     2.740
     1.830
     2.360
     2.460


y:
  27.00000
  57.00000
  98.00000
 267.00000
  59.00000
 142.00000
 110.00000
 135.00000
  91.00000
 119.00000
  62.00000
  40.00000
 166.00000
 116.00000
 335.00000
  39.00000
  67.00000
  48.00000
  35.00000
  33.00000
  48.00000
  35.00000

这是我的损失和重量：

loss weights