diff --git "a/docs/16.\345\274\272\345\214\226\345\255\246\344\271\240.md" "b/docs/16.\345\274\272\345\214\226\345\255\246\344\271\240.md"
index 10bc836c5c18f4fdf809c80cc48cba936927d6ba..33f5958208839eceb63bc7f3708528011312f3ac 100644
--- "a/docs/16.\345\274\272\345\214\226\345\255\246\344\271\240.md"
+++ "b/docs/16.\345\274\272\345\214\226\345\255\246\344\271\240.md"
@@ -169,7 +169,8 @@ n_hidden = 4  # 这只是个简单的测试，不需要过多的隐藏层
 n_outputs = 1 # 只输出向左加速的概率
 initializer = tf.contrib.layers.variance_scaling_initializer()
 # 2. 建立神经网络 
-X = tf.placeholder(tf.float32, shape=[None, n_inputs]) hidden = fully_connected(X, n_hidden, activation_fn=tf.nn.elu,weights_initializer=initializer) # 隐层激活函数使用指数线性函数                
+X = tf.placeholder(tf.float32, shape=[None, n_inputs])
+hidden = fully_connected(X, n_hidden, activation_fn=tf.nn.elu,weights_initializer=initializer) # 隐层激活函数使用指数线性函数             
 logits = fully_connected(hidden, n_outputs, activation_fn=None,weights_initializer=initializer)                   
 outputs = tf.nn.sigmoid(logits)
 # 3. 在概率基础上随机选择动作