📄 neural_network.cpp
字号:
}
}
/*隐藏层和隐藏层之间的权重更新*/
for(i=0;i<GetNum_h()-1;i++)
{
for(j=0;j<h_layers[i].GetNode_Num();j++)
{
for(k=0;k<h_layers[i+1].GetNode_Num();k++)
{
m_w = wp[m_w_id].GetW_w() + l*(h_layers[i+1].GetNode(k).GetErr())*(h_layers[i].GetNode(j).GetO_value());
wp[m_w_id].SetW_w(m_w);
m_w_id++;
}
}
}
/*隐藏层和输出层之间的权重更新*/
for(j=0;j<h_ln_num;j++)
{
for(i=0;i<o_n_num;i++)
{
m_w = wp [m_w_id].GetW_w() + l*(o_layers[0].GetNode(i).GetErr())*(h_layers[GetNum_h()-1].GetNode(j).GetO_value()); //按公式计算
wp[m_w_id].SetW_w(m_w);
m_w_id++;
}
}
/*隐藏层和输出层节点的偏倚更新*/
/*隐藏层*/
for(i=0;i<GetNum_h();i++)
{
for(j=0;j<h_layers[i].GetNode_Num();j++)
{
m_bas = h_layers[i].GetNode(j).GetBias() + l*(h_layers[i].GetNode(j).GetErr());
h_layers[i].GetNode(j).SetBias(m_bas);
}
}
/*输出层*/
for(j=0;j<o_n_num;j++)
{
m_bas = o_layers[0].GetNode(j).GetBias() + l*(o_layers[0].GetNode(j).GetErr()); //按公式计算
o_layers[0].GetNode(j).SetBias(m_bas);
}
} //周期结束
p_data = temp;
err_kind = err_kind_num/num_data;
if(err_kind<m_err_kind)
{
isStop = true;
}
}
}
CString CNeural_NetWork::GetResult()
{
/*输出训练停止后的各节点权重和偏倚*/
CString outstr = "";
CString tempstr = "";
CString cstr = "";
float *temp_hbas = new float[count_of_hbas];
float *temp_obas = new float[o_layers[0].GetNode_Num()];
int index = 0;
for(int i=0;i<GetNum_h();i++)
{
for(int j=0;j<h_layers[i].GetNode_Num();j++)
{
tempstr.Format("%d",index+1);
cstr.Format("%f",h_layers[i].GetNode(j).GetBias());
outstr = outstr+"隐藏层结点"+tempstr+"偏倚:"+cstr+"\r\n";
temp_hbas[index] = h_layers[i].GetNode(j).GetBias();
index++;
}
}
for(i=0;i<o_layers[0].GetNode_Num();i++)
{
tempstr.Format("%d",i+1);
cstr.Format("%f",o_layers[0].GetNode(i).GetBias());
outstr = outstr+"输出层结点"+tempstr+"偏倚:"+cstr+"\r\n";
temp_obas[i] = o_layers[0].GetNode(i).GetBias();
}
tempstr = "";
cstr.Format("%f",o_layers[0].GetNode(0).GetO_value());
tempstr = outstr+"结点输出值:"+cstr+"\r\n";
cstr.Format("%f",o_layers[0].GetNode(0).GetErr());
tempstr = tempstr + "输出结点误差:"+cstr;
return tempstr;
}
void CNeural_NetWork::ParametersSave(int n)
{
BOOL si,so,sh,shbas,sobas,sw,sh_l;
CString tempstr = "";
CString outstr = "";
CString cstr = "";
int isSave=AfxMessageBox("保存会覆盖上一次的数据,确认保存?",MB_YESNO|MB_ICONINFORMATION,-1);
if(isSave == 6)
{
cstr.Format("%d",o_layers[0].GetNode_Num());
so = WriteToSaveFile(cstr,"NumOfOlayerNodes.txt"); //保存输出层结点数
cstr.Format("%d",i_layers[0].GetNode_Num());
si = WriteToSaveFile(cstr,"NumOfIlayerNodes.txt"); //保存输入层结点数
cstr.Format("%d",GetNum_h());
sh_l = WriteToSaveFile(cstr,"NumOfHlayer.txt"); //保存隐藏层数
for(int i=0;i<GetNum_h();i++)
{
tempstr.Format("%d",h_layers[i].GetNode_Num());
outstr = outstr + tempstr +"\r\n";
}
sh = WriteToSaveFile(outstr,"NumOfHlayerNodes.txt"); //保存隐藏层各层的结点数
outstr = ""; //清空
for(i=0;i<GetNum_h();i++)
{
for(int j=0;j<h_layers[i].GetNode_Num();j++)
{
tempstr.Format("%f",h_layers[i].GetNode(j).GetBias());
outstr = outstr + tempstr +"\r\n";
}
}
shbas = WriteToSaveFile(outstr,"HlayesBas.txt"); //保存隐藏层各层的结点的偏倚
outstr = "";
for(i=0;i<o_layers[0].GetNode_Num();i++)
{
tempstr.Format("%f",o_layers[0].GetNode(i).GetBias());
outstr = outstr + tempstr +"\r\n";
}
sobas = WriteToSaveFile(outstr,"OlayesBas.txt"); //保存输出层各层的结点的偏倚
outstr = "";
for (i=0;i<n;i++)
{
tempstr.Format("%f",wp[i].GetW_w());
outstr = outstr + tempstr +"\r\n";
}
sw = WriteToSaveFile(outstr,"w.txt"); //保存网络权重
if(si&&so&&sh&&shbas&&sobas&&sw&&sh_l)
{
AfxMessageBox("保存成功!");
return;
}
else
{
AfxMessageBox("保存失败!");
return;
}
}
}
BOOL CNeural_NetWork::WriteToSaveFile(CString m_str,CString FileName)
{
CStdioFile file;
file.Open(FileName,CFile::modeCreate|CFile::modeReadWrite);
file.WriteString(m_str);
file.Close();
return TRUE;
}
int CNeural_NetWork::ReadFromSaveFile(CString FileName)
{
CStdioFile file;
CString tempstr;
int thedata;
file.Open(FileName,CFile::modeRead|CFile::typeText);
while(file.ReadString(tempstr))
{
tempstr.TrimLeft();
tempstr.TrimRight();
thedata = int(atoi(tempstr));
}
file.Close();
return thedata;
}
float* CNeural_NetWork::ReadPFromSaveFile(CString FileName)
{
CStdioFile file;
CString tempstr;
int NumofIndex = 0;
file.Open(FileName,CFile::modeRead|CFile::typeText);
while(file.ReadString(tempstr))
{
NumofIndex++;
}
file.Close();
tempstr = "";
float *m_p_data = new float[NumofIndex];
NumofIndex = 0;
file.Open(FileName,CFile::modeRead|CFile::typeText);
while(file.ReadString(tempstr))
{
tempstr.TrimLeft();
tempstr.TrimRight();
m_p_data[NumofIndex++] = float(atof(tempstr));
}
file.Close();
return m_p_data;
}
CWeight* CNeural_NetWork::ReadWFromSaveFile(CString FileName)
{
CStdioFile file;
CString tempstr;
int NumofIndex = 0;
file.Open(FileName,CFile::modeRead|CFile::typeText);
while(file.ReadString(tempstr))
{
NumofIndex++;
}
file.Close();
tempstr = "";
CWeight *m_p_data = new CWeight[NumofIndex];
NumofIndex = 0;
file.Open(FileName,CFile::modeRead|CFile::typeText);
while(file.ReadString(tempstr))
{
tempstr.TrimLeft();
tempstr.TrimRight();
m_p_data[NumofIndex++].SetW_w(float(atof(tempstr)));
}
file.Close();
return m_p_data;
}
int* CNeural_NetWork::ReadPHFromSaveFile(CString FileName)
{
CStdioFile file;
CString tempstr;
int NumofIndex = 0;
file.Open(FileName,CFile::modeRead|CFile::typeText);
while(file.ReadString(tempstr))
{
NumofIndex++;
}
file.Close();
tempstr = "";
int *m_p_data = new int[NumofIndex];
NumofIndex = 0;
file.Open(FileName,CFile::modeRead|CFile::typeText);
while(file.ReadString(tempstr))
{
tempstr.TrimLeft();
tempstr.TrimRight();
m_p_data[NumofIndex++] = float(atof(tempstr));
}
file.Close();
return m_p_data;
}
float CNeural_NetWork::GetOvalue(float *p_data)
{
int i,j,k,m,n,t; //循环计数变量
float i_value,o_value; //临时变量用于存储节点净输入和净输出
float err; //临时变量存储节点误差
float t_value; //临时变量存储输出节点期望
short h_n_num,o_n_num,i_n_num,h_ln_num,h_an_num=0; //分别存储隐藏层第一层节点数和输出层节点数,输入层节点数,
//隐藏层最后一层节点数,隐藏层总的节点数
short ih_w_num =0,hh_w_num = 0; //ih_w_num,输入层到隐藏层第一层的权重总数,hh_w_num隐藏层之间的权重总数
short m_w_id; //权重主标识
float m_w; //临时变量存储权重
float m_bas; //临时变量存储偏倚
int idx = 0; //数据数组的行索引
double train_t_num=1;
isStop = false;
h_n_num = h_layers[0].GetNode_Num();
o_n_num = o_layers[0].GetNode_Num();
i_n_num = i_layers[0].GetNode_Num();
h_ln_num = h_layers[GetNum_h()-1].GetNode_Num();
for(j=0;j<GetNum_h();j++)
{
h_an_num = h_an_num + h_layers[j].GetNode_Num();
}
ih_w_num = i_layers[0].GetNode_Num()*h_layers[0].GetNode_Num();
for(int ii_h=0;ii_h<GetNum_h()-1;ii_h++)
{
hh_w_num = hh_w_num+h_layers[ii_h].GetNode_Num()*h_layers[ii_h+1].GetNode_Num();
}
float w_sum = 0; //w_sum存储节点加权和
float *temp;
temp = p_data;
/*算法开始*/
m_w_id = 0; //权重主标识清零
for(i=0;i<i_n_num;i++)
{
i_layers[0].GetNode(i).SetI_value(*p_data); //设置元组输入层单元的输入值
p_data++;
}
for(i=0;i<o_n_num;i++)
{
o_layers[0].GetNode(i).SetT(*p_data); //设置元组输出层单元对应期望
p_data++;
}
/*向前传播输入*/
for(i=0;i<h_n_num;i++)
{
for(j=0;j<i_n_num;j++)
{
k = j*h_n_num;
w_sum = w_sum+(wp[k+i].GetW_w())*(i_layers[0].GetNode(j).GetI_value());
}
i_value = w_sum+h_layers[0].GetNode(i).GetBias(); //计算出隐藏层第一层各个节点的净输入
o_value = 1/(1+exp(-i_value)); //计算出隐藏层第一层各个节点的净输出
h_layers[0].GetNode(i).SetI_value(i_value);
h_layers[0].GetNode(i).SetO_value(o_value);
w_sum = 0;
}
/////////////////////////////////////////////////////
n = h_n_num*i_n_num;
for(i=0;i<GetNum_h()-1;i++)
{
for(j=0;j<h_layers[i+1].GetNode_Num();j++)
{
for(k=0;k<h_layers[i].GetNode_Num();k++)
{
m = k*h_layers[i+1].GetNode_Num();
w_sum = w_sum + (wp[m+j+n].GetW_w())*(h_layers[i].GetNode(k).GetO_value());
}
i_value = w_sum+h_layers[i+1].GetNode(j).GetBias(); //隐藏层其它层净输入
o_value = 1/(1+exp(-i_value)); //隐藏层其它层净输出
h_layers[i+1].GetNode(j).SetI_value(i_value);
h_layers[i+1].GetNode(j).SetO_value(o_value);
w_sum = 0;
}
n = n + h_layers[i].GetNode_Num()*h_layers[i+1].GetNode_Num();
}
/////////////////////////////////////////////////////
for(i=0;i<o_n_num;i++)
{
for(j=0;j<h_ln_num;j++)
{
k = j*o_n_num;
w_sum = w_sum+(wp[(k+i)+(ih_w_num+hh_w_num)].GetW_w())*(h_layers[GetNum_h()-1].GetNode(j).GetO_value());
}
i_value = w_sum+o_layers[0].GetNode(i).GetBias(); //计算出输出层各个节点的净输入
o_value = 1/(1+exp(-i_value)); //计算出输出层各个节点的净输出
o_layers[0].GetNode(i).SetI_value(i_value);
o_layers[0].GetNode(i).SetO_value(o_value);
w_sum = 0;
}
return o_layers[0].GetNode(0).GetO_value();
}
CNeural_NetWork::~CNeural_NetWork()
{
delete [] h_layers;
delete [] o_layers;
delete [] i_layers;
}
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -