深度學習全連線的程式碼如下：

• #-*-coding:utf-8-*-
  
  #匯入所需要的庫
  import logging
  import math
  import random
  import mxnet as mx
  import numpy as np
  
  logging.getLogger().setLevel(logging.DEBUG) #開啟除錯資訊的顯示
  
  n_sample = 10000 #訓練用的資料點個數
  batch_size = 10 #批大小
  learning_rate = 0.1 #學習速率
  n_epoch = 20 #訓練epoch數
  
  #每組資料都是在（0，1）之間的兩個隨機數
  train_in = [[ random.uniform(0,1) for c in range(2)] for n in range(n_sample)]
  train_out = [0 for n in range(n_sample)] #期望輸出，先初始化為0
  
  #每組資料的期望輸出是其中的大者
  for i in range(n_sample):
      train_out[i] = max(train_in[i][0], train_in[i][1])
  
  #訓練資料的迭代器
  train_iter = mx.io.NDArrayIter(data = np.array(train_in), label = {'reg_label':np.array(train_out)}, batch_size = batch_size, shuffle = True)
  
  src = mx.sym.Variable('data') #輸入層
  fc1 = mx.sym.FullyConnected(data = src, num_hidden = 10, name = 'fc1') #全連線層
  act1 = mx.sym.Activation(data = fc1, act_type = "relu", name = 'act1') #ReLU連線層
  fc2 = mx.sym.FullyConnected(data = act1, num_hidden = 10, name = 'fc2') #全連線層
  act2 = mx.sym.Activation(data = fc2, act_type = "relu", name = 'act2') #ReLU連線層
  fc3 = mx.sym.FullyConnected(data = act2, num_hidden = 1, name = 'fc3') #全連線層
  net = mx.sym.LinearRegressionOutput(data = fc3, name = 'reg') #輸出層
  
  #完成網路定義
  module = mx.mod.Module(symbol = net, label_names = (['reg_label']))
  
  #訓練語句：
  module.fit(
      train_iter, #訓練資料的迭代器
      eval_data = None, #在此只訓練，不使用測試資料
      eval_metric = mx.metric.create('mse'), #輸出MSE損失資訊
      initializer = mx.initializer.Uniform(0.5), #將權重和偏置設定在[-0.5,0.5]之間均勻的隨機數
      optimizer = 'sgd', #梯度下降演算法為SGD
      optimizer_params = {'learning_rate': learning_rate}, #設定學習速率
      num_epoch = n_epoch, #訓練epoch數
  
      batch_end_callback = None, #減少輸出資訊
      epoch_end_callback = None #減少輸出資訊
  )
   
  

  輸出部分如下：

• 其中Train-mse是損失資訊佔總的百分比

• 可調節訓練次數n_epoch，以此來調節輸出。由以下輸出可知，當訓練次數n_epoch=14時，損失就不下降，恆為0.000001。即此訓練模型的擬合達到了99.9999%，接近於100%。讀者可調節訓練次數，可以發現，當訓練次數越多，損失值就越低。

• INFO:root:Epoch[0] Train-mse=0.019081
  INFO:root:Epoch[0] Time cost=2.080
  INFO:root:Epoch[1] Train-mse=0.009061
  INFO:root:Epoch[1] Time cost=1.824
  INFO:root:Epoch[2] Train-mse=0.002438
  INFO:root:Epoch[2] Time cost=1.848
  INFO:root:Epoch[3] Train-mse=0.000335
  INFO:root:Epoch[3] Time cost=1.915
  INFO:root:Epoch[4] Train-mse=0.000185
  INFO:root:Epoch[4] Time cost=1.643
  INFO:root:Epoch[5] Train-mse=0.000087
  INFO:root:Epoch[5] Time cost=2.048
  INFO:root:Epoch[6] Train-mse=0.000039
  INFO:root:Epoch[6] Time cost=2.187
  INFO:root:Epoch[7] Train-mse=0.000020
  INFO:root:Epoch[7] Time cost=2.166
  INFO:root:Epoch[8] Train-mse=0.000011
  INFO:root:Epoch[8] Time cost=2.325
  INFO:root:Epoch[9] Train-mse=0.000006
  INFO:root:Epoch[9] Time cost=1.985
  INFO:root:Epoch[10] Train-mse=0.000004
  INFO:root:Epoch[10] Time cost=1.874
  INFO:root:Epoch[11] Train-mse=0.000003
  INFO:root:Epoch[11] Time cost=1.777
  INFO:root:Epoch[12] Train-mse=0.000002
  INFO:root:Epoch[12] Time cost=1.783
  INFO:root:Epoch[13] Train-mse=0.000002
  INFO:root:Epoch[13] Time cost=1.874
  INFO:root:Epoch[14] Train-mse=0.000001
  INFO:root:Epoch[14] Time cost=1.791
  INFO:root:Epoch[15] Train-mse=0.000001
  INFO:root:Epoch[15] Time cost=2.504
  INFO:root:Epoch[16] Train-mse=0.000001
  INFO:root:Epoch[16] Time cost=1.733
  INFO:root:Epoch[17] Train-mse=0.000001
  INFO:root:Epoch[17] Time cost=1.620
  INFO:root:Epoch[18] Train-mse=0.000001
  INFO:root:Epoch[18] Time cost=1.931
  INFO:root:Epoch[19] Train-mse=0.000001
  INFO:root:Epoch[19] Time cost=2.272