[AI]人工神经网络对葡萄酒打分

人工神经网络（artificial neural network，缩写ANN），简称神经网络（neural network，缩写NN）或类神经网络，是一种模仿生物神经网络(动物的中枢神经系统，特别是大脑)的结构和功能的数学模型或计算模型。神经网络由大量的人工神经元联结进行计算。大多数情况下人工神经网络能在外界信息的基础上改变内部结构，是一种自适应系统。现代神经网络是一种非线性统计性数据建模工具，常用来对输入和输出间复杂的关系进行建模，或用来探索数据的模式。（摘自https://zh.wikipedia.org/wiki/%E4%BA%BA%E5%B7%A5%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C）

给定葡萄酒的各项属性及评分，训练一个人工神经网络，对任意给定新的葡萄酒进行打分。

实验数据如下：

样品编号	蛋白质mg/100g（x1）	DPPH自由基1/IC50（g/L）（x2）	总酚(mmol/kg)（x3）	葡萄总黄酮（mmol/kg）（x4）	PH值（x5）	果皮质量（g）（x6）	总分（y）
葡萄样品1	555.455	0.4314	23.576	9.509	3.54	0.120	77.1
葡萄样品2	624.094	0.4659	26.026	13.720	3.88	0.193	78.2
葡萄样品3	580.273	0.4102	21.479	10.853	3.80	0.160	74.6
葡萄样品4	527.438	0.2660	10.783	4.394	3.36	0.173	75.8
葡萄样品5	590.651	0.3972	18.547	10.333	3.58	0.260	74
葡萄样品6	532.026	0.2755	10.469	6.867	3.31	0.213	74.5
葡萄样品7	489.320	0.1758	9.181	3.497	3.13	0.136	72.6
葡萄样品8	556.091	0.4160	15.343	8.454	2.90	0.240	71.5
葡萄样品9	703.300	0.6689	31.767	20.433	3.68	0.150	72.2
葡萄样品10	547.695	0.3263	9.191	4.603	3.66	0.210	71.6
葡萄样品11	545.034	0.2796	6.197	2.545	3.46	0.125	72
葡萄样品12	491.265	0.1975	11.924	3.926	3.37	0.253	72.6
葡萄样品13	603.686	0.4420	14.572	7.360	3.91	0.170	72.1
葡萄样品14	597.274	0.3606	15.661	7.780	3.46	0.256	69.9
葡萄样品15	531.431	0.2193	12.001	5.598	3.16	0.208	71.5
葡萄样品16	585.783	0.2371	10.992	9.185	3.25	0.138	68.8
葡萄样品17	546.516	0.3594	15.394	8.613	3.38	0.336	68.8
葡萄样品18	511.565	0.2260	7.979	5.273	3.33	0.150	71.2
葡萄样品19	542.201	0.3806	16.732	9.370	3.69	0.089	66.3
葡萄样品20	556.895	0.2825	11.914	8.069	3.66	0.247	66
葡萄样品21	566.188	0.3803	15.639	7.542	3.66	0.107	68.2
葡萄样品22	490.759	0.2842	16.066	7.759	3.72	0.126	65.3
葡萄样品23	547.813	0.5748	32.522	24.436	3.42	0.263	61.6
葡萄样品24	504.429	0.2836	8.192	8.265	3.65	0.227	68.1
葡萄样品25	539.502	0.3518	11.813	5.429	3.41	0.201	65.4
葡萄样品26	589.906	0.3179	7.129	3.413	3.63	0.210	65.7
葡萄样品27	523.837	0.2654	9.145	4.711	3.31	0.189	68.3

由于样本不足，存在过拟合等问题

源码：

#include <iostream>
#include <cstdlib>
#include <cstring>
#include <fstream>
#include <cmath>
#include <ctime>
#define HIDENODE 5
#define RATE 0.9
#define TOLATHRE 0.001
#define PRECISION 10000
#define TOTALNUM 27
#define SAMPLENUM 17
#define ATTRNUM 7
using namespace std;
double SampleData[SAMPLENUM][ATTRNUM];
double TotalData[TOTALNUM][ATTRNUM], EvalData[TOTALNUM], MaxEval, MinEval;
double hi[HIDENODE], ho[HIDENODE], yi[SAMPLENUM], yo[SAMPLENUM];
double wih[HIDENODE][ATTRNUM - 1], who[HIDENODE];
double bh[HIDENODE], bo;
double Distin = 0x7FFFFFFF;
inline double f(double x)
{
return 1.0 / (1 + exp(-1 * x));
}

void Init()
{
freopen("ANN.txt", "r", stdin);
double max_attr[ATTRNUM], min_attr[ATTRNUM];
memset(max_attr, 0, sizeof(max_attr));
for (int i = 0; i< ATTRNUM; i++)
min_attr[i] = 0x7FFFFFFF;
//Input data
for (int i = 0; i < SAMPLENUM; i++)
{
for (int j = 0; j < ATTRNUM; j++)
{
cin >> SampleData[i][j];
max_attr[j] = (max_attr[j]>SampleData[i][j]) ? max_attr[j] : SampleData[i][j];
min_attr[j] = (min_attr[j]<SampleData[i][j]) ? min_attr[j] : SampleData[i][j];
}
}
//Normalize data
for (int i = 0; i < ATTRNUM; i++)
{
double diff = max_attr[i] - min_attr[i];
for (int j = 0; j < SAMPLENUM; j++)
SampleData[j][i] = (SampleData[j][i] - min_attr[i]) / diff;
}
//Rand weight
for (int i = 0; i < HIDENODE; i++)
{
who[i] = rand() % (2 * PRECISION + 1) / (double)PRECISION - 1;
for (int j = 0; j < ATTRNUM - 1; j++)
wih[i][j] = rand() % (2 * PRECISION + 1) / (double)PRECISION - 1;
}
for (int i = 0; i < SAMPLENUM; i++)
bh[i] = rand() % (2 * PRECISION + 1) / (double)PRECISION - 1;
bo = rand() % (2 * PRECISION + 1) / (double)PRECISION - 1;
}
void Train()
{
double theta1[HIDENODE], theta2 = 0;
Distin = 0;
//cout << Distin << ' ';
for (int i = 0; i < SAMPLENUM; i++)
{
memset(hi, 0, sizeof(hi));
memset(theta1, 0, sizeof(theta1));
yi[i] = 0;
for (int j = 0; j < HIDENODE; j++)
for (int k = 0; k < 6; k++)
hi[j] += SampleData[i][k] * wih[j][k];
for (int j = 0; j < HIDENODE; j++)
ho[j] = f(hi[j] + bh[j]);
for (int j = 0; j < HIDENODE; j++)
yi[i] += who[j] * ho[j];
yo[i] = f(yi[i] + bo);
theta2 = (SampleData[i][ATTRNUM - 1] - yo[i])*yo[i] * (1 - yo[i]);
for (int j = 0; j < HIDENODE; j++)
who[j] += RATE*theta2*ho[j];

for (int j = 0; j < HIDENODE; j++)
{
theta1[j] += theta2 * who[j] * ho[j] * (1 - ho[j]);
for (int k = 0; k < 6; k++)
wih[j][k] += RATE*theta1[j] * SampleData[i][k];
}
bo += RATE*theta2;
for (int j = 0; j < HIDENODE; j++)
bh[j] += RATE*theta1[j];

// compute Distin
Distin += pow(yo[i] - SampleData[i][6], 2);
}
//cout << Distin << endl;
}
void OutputParaments()
{
cout << "weights between input layer and hidden layer:" << endl;
for (int i = 0; i < HIDENODE; i++)
{
for (int j = 0; j < ATTRNUM - 1; j++)
cout << wih[i][j] << ' ';
cout << endl;
}

cout << endl << "weights between hidden layer and output layer:" << endl;
for (int i = 0; i < HIDENODE; i++)
cout << who[i] << ' ';
cout << endl;

cout << endl << "Thresholds in hidden layer:" << endl;
for (int i = 0; i < HIDENODE; i++)
cout << bh[i] << ' ';
cout << endl << endl;

cout << "Threshold in output layer:" << endl << bo << endl << endl;
}
void CalculateDifferences()
{
double max_diff = 0, min_diff = 0x7FFFFFFF, aver_diff = 0;
for (int i = 0; i < SAMPLENUM; i++)
{
aver_diff += fabs(SampleData[i][ATTRNUM - 1] - yo[i]);
max_diff = (max_diff > fabs(SampleData[i][ATTRNUM - 1] - yo[i])) ? max_diff : fabs(SampleData[i][ATTRNUM - 1] - yo[i]);
min_diff = (min_diff < fabs(SampleData[i][ATTRNUM - 1] - yo[i])) ? min_diff : fabs(SampleData[i][ATTRNUM - 1] - yo[i]);
}
aver_diff /= SAMPLENUM;
cout << "Max difference is: " << max_diff << endl;
cout << "Min difference is: " << min_diff << endl;
cout << "Average difference is: " << aver_diff << endl;
}
void GetAllGrap()
{
ifstream fin("ANN.txt");
double max_attr[ATTRNUM], min_attr[ATTRNUM];
memset(max_attr, 0, sizeof(max_attr));
MaxEval = 0;
MinEval = 0x7FFFFFFF;
for (int i = 0; i< ATTRNUM; i++)
min_attr[i] = 0x7FFFFFFF;
for (int i = 0; i < TOTALNUM; i++)
{
for (int j = 0; j < ATTRNUM; j++)
{
fin >> TotalData[i][j];
max_attr[j] = (max_attr[j]>TotalData[i][j]) ? max_attr[j] : TotalData[i][j];
min_attr[j] = (min_attr[j]<TotalData[i][j]) ? min_attr[j] : TotalData[i][j];
}
EvalData[i] = TotalData[i][ATTRNUM - 1];
MaxEval = (MaxEval>TotalData[i][ATTRNUM - 1]) ? MaxEval : TotalData[i][ATTRNUM - 1];
MinEval = (MinEval<TotalData[i][ATTRNUM - 1]) ? MinEval : TotalData[i][ATTRNUM - 1];
}
for (int i = 0; i < ATTRNUM; i++)
{
double diff = max_attr[i] - min_attr[i];
for (int j = 0; j < TOTALNUM; j++)
TotalData[j][i] = (TotalData[j][i] - min_attr[i]) / diff;
}
fin.close();
}
void CalculateEval(int grapnum)
{
double tmp_hidden[HIDENODE];
memset(tmp_hidden, 0, sizeof(tmp_hidden));
for (int i = 0; i < HIDENODE; i++)
{
for (int j = 0; j < ATTRNUM - 1; j++)
{
tmp_hidden[i] += TotalData[grapnum][j] * wih[i][j];
}
tmp_hidden[i] -= bh[i];
}
double tmp_output = 0;
for (int i = 0; i < HIDENODE; i++)
{
for (int j = 0; j < ATTRNUM - 1; j++)
tmp_output += tmp_hidden[i] * who[i];
}
tmp_output -= bo;
//tmp_output*=(MaxEval - MinEval) + MinEval;
cout << "the test result is: " << tmp_output << endl;
cout << "the distinguish is: " << fabs(tmp_output-EvalData[grapnum])<<endl;
}
int main()
{
srand((unsigned)time(0));
memset(hi, 0, sizeof(hi));
memset(ho, 0, sizeof(ho));
memset(yi, 0, sizeof(yi));
memset(yo, 0, sizeof(yo));
Init();
while (Distin > TOLATHRE)
Train();
OutputParaments();
CalculateDifferences();
int TestGrap = 0;
GetAllGrap();
freopen("con", "r", stdin);
cout<<"input the test grap num:"<<endl;
while (cin >> TestGrap)
{
if(TestGrap>=0&&TestGrap<=26)
{
CalculateEval(TestGrap);
cout<<"input the test grap num:"<<endl;
}
else
cout<<"Not legal!"<<endl;
}
return 0;
}