神经网络: 计算简单的加法

神经网络: 计算简单的加法

下面是完整的代码:

1. //
2. //fann_test.c:
3. //FANN测试
4. //
5. //到下面的网站下载FANN库:
6. //FastArtificialNeuralNetworkLibrary(fann)
7. //http://leenissen.dk/fann/
8. //下载链接:DownloadFANN--->CSourceCodeandWindowsDLLfiles
9. //
11. #include"../fann-2.0.0/src/include/doublefann.h"
13. #ifdef_DEBUG
14. #pragmacomment(lib,"../fann-2.0.0/MicrosoftWindowsDll/bin/fanndoubleMTd.lib")
15. #else
16. #pragmacomment(lib,"../fann-2.0.0/MicrosoftWindowsDll/bin/fanndoubleMT.lib")
17. #endif
19. //训练:
20. //加法神经网络
21. //c=a+b;
22. voidtrain()
23. {
24. constunsignedintnum_input=2;//输入项个数
25. constunsignedintnum_output=1;//输出项个数
26. constunsignedintnum_layers=3;
27. constunsignedintnum_neurons_hidden=3;
28. constfloatdesired_error=(constfloat)0.00000001;
29. constunsignedintmax_epochs=500000;//最多执行次数
30. constunsignedintepochs_between_reports=10000;//报告频率
32. structfann*ann;
34. intNum=200;
35. floatMf=Num*3.f;
36. inti;
37. doublea,b,c;
39. FILE*fp;
40. fopen_s(&fp,"add.fann","w");
41. fprintf_s(fp,"%d21/n",Num);
43. //生成训练文件
44. for(i=1;i<=Num;i++){
45. //生成2个数,要求在(0,1)之间
46. a=i/Mf;
47. b=(i+1)/Mf;
48. c=a+b;//要求在(0,1)之间
50. //输入内容写到训练文件
51. fprintf_s(fp,"%lf%lf/n%lf/n",a,b,c);
52. }
53. fclose(fp);
55. //样本训练
56. ann=fann_create_standard(num_layers,num_input,num_neurons_hidden,num_output);
58. fann_set_activation_function_hidden(ann,FANN_LINEAR);
59. fann_set_activation_function_output(ann,FANN_LINEAR);
61. fann_train_on_file(ann,"add.fann",max_epochs,epochs_between_reports,desired_error);
62. fann_save(ann,"add.fann.net");
63. fann_destroy(ann);
64. }
66. //执行:
67. //测试
68. voidexec(doublea,doubleb)
69. {
70. structfann*ann;
72. fann_type*calc_out;
73. fann_typeinput[2];
74. ann=fann_create_from_file("add.fann.net");
76. input[0]=a;
77. input[1]=b;
79. calc_out=fann_run(ann,input);
80. fann_destroy(ann);
82. printf("a=%f/nb=%f/nc=%f/n期望值c=%f/n/n",input[0],input[1],calc_out[0],input[0]+input[1]);
83. }
85. //
86. //主程序
87. //
88. intmain()
89. {
90. //下面的方法只需调用一次,然后注释掉
91. //train();
93. exec(0.354,0.58934);
94. exec(0.21469,0.3914968);
95. exec(0.130,0.44);
96. exec(-0.3654,0.58455);
97. exec(0.365420,-0.95);
99. return0;
100. }