public void CLEAR_STOCK(ref stock_[] bp_stock, int kind)
{
stock_[] clear_stock = new stock_[Global_days._days];
clear_stock = bp_stock;
for (int i = 0; i < Global_days._days; i++)
{
clear_stock[i].s_date = 0;
clear_stock[i].s_dcp_int = 0;
clear_stock[i].s_dhp_int = 0;
clear_stock[i].s_dlp_int = 0;
clear_stock[i].s_dmp_int = 0;
clear_stock[i].s_dtv_int = 0;
}
}
/*BPA Learn*/
//1.시가 2.고가 3.저가 4.거래량 5.종가(타겟)
public double BP_START_STOCK(ref stock_[] bp_stock, int kind)
{
//Initial
stock_[] stock_bp = new stock_[Global_days._days];
stock_bp = bp_stock; //ref
switch (kind)
{
case Kind_Constants.test_60days:
Input = new double[Get_60days * Input_Neuron];
break;
case Kind_Constants.test_120days:
Input = new double[Get_120days * Input_Neuron];
break;
}
//Bias, Hidden Neuron Weight Set
for (int i = 0; i < Number_Neurons; i++)
{
Bias_Weight[i] = ran.NextDouble();
}
for (int i = 0; i < Input_Neuron * Hd_L_Number; i++)
{
Weight_Input_Layer[i] = ran.NextDouble();
}
for (int i = 0; i < Number_Layer; i++)
{
for (int j = 0; j < Hd_L_Number * Hd_L_Number; j++)
{
Weight_Layer[i, j] = ran.NextDouble();
}
}
for (int i = 0; i < Output_Neuron * Hd_L_Number; i++)
{
Weight_Output_Layer[i] = ran.NextDouble();
}
//max count get
for (int i = 0; i < Get_days; i++)
{
if (max_count < stock_bp[i].s_dhp_int)
{
max_count = stock_bp[i].s_dhp_int;
}
}
do
{
max_count = (max_count / 10);
digits++;
} while (max_count > 0);
for (int i = 0; i < Get_days; i++)
{
if (max_count_tv < stock_bp[i].s_dtv_int)
{
max_count_tv = stock_bp[i].s_dtv_int;
}
}
do
{
max_count_tv = (max_count_tv / 10);
digits_tv++;
} while (max_count_tv > 0);
//Input Set
for (int i = 0; i < Get_days; i++)
{
Input[i + 0] = stock_bp[i].s_dmp_int;
Input[i + 1] = stock_bp[i].s_dhp_int;
Input[i + 2] = stock_bp[i].s_dlp_int;
Input[i + 3] = stock_bp[i].s_dtv_int;
Input[i + 0] /= Math.Pow(10, digits);
Input[i + 1] /= Math.Pow(10, digits);
Input[i + 2] /= Math.Pow(10, digits);
Input[i + 3] /= Math.Pow(10, digits_tv);
}
for (int i = 0; i < Get_days; i++)
{
for (int j = 0; j < Output_Neuron; j++)
{
Target_t[i, j] = stock_bp[i].s_dcp_int;
Target_t[i, j] /= Math.Pow(10, digits);
}
}
//Output Set
//BPA Start
while (Epoch-- > 0)
{
/*Input - Hidden Layer[0] 사이 Sum,Sigmoid,Delta */
for (int i = 0; i < Input_Neuron; i++)
{
for (int j = 0; j < Input_Neuron; j++)
{
Sum[i] += Input[j + bnc * Input_Neuron] * Weight_Input_Layer[inc];
++inc;
}
Sum[i] += (Bias * Bias_Weight[i]);
Sigmoid[i] = (1.0 / (1.0 + Math.Exp(-Sum[i])));
}
inc = 0;
/*Hidden Layer 사이의 Sum, Sigmoid*/
for (int i = Number_Layer - 1; i > 0; i--)
{
k += Hd_L_Number;
//ex) 20,21,22,23,24 / 15,16,17,18,19 / ...
for (int j = New_Lable - (Hd_L_Number + jump); j < New_Lable - jump; j++)
{
//ex) 25-(5+5*k) -> n=20-5k; n < 25-5k; n++ -> 20,21,22,23,24 / 15,16,17,18,19 / ....
for (int n = New_Lable - (Hd_L_Number + k); n < New_Lable - k; n++)
{
Sum[j] += (Sigmoid[n] * Weight_Layer[i - 1, inc]);
++inc;
}
Sum[j] += (Bias * Bias_Weight[j]);
Sigmoid[j] = (1.0 / (1.0 + Math.Exp(-Sum[j])));
}
inc = 0;
jump += Hd_L_Number;
}
jump = 0;
k = 0;
/* Output Layer와 연결된 Hidden Layer이용하여 Output Sum,Sigmoid */
for (int i = 0; i < Output_Neuron; ++i)
{
for (int j = Lable; j < New_Lable; j++)
{
Sum_Output[i] += (Sigmoid[j] * Weight_Output_Layer[inc]);
inc++;
}
Sum_Output[i] += (Bias * Bias_Weight[New_Lable + i]);
Sigmoid_Output[i] = (1.0 / (1.0 + Math.Exp(-Sum_Output[i])));
Delta_Output[i] = (Sigmoid_Output[i] * (1 - Sigmoid_Output[i])) * (Target_t[bnc, i] - Sigmoid_Output[i]);
/*Target 값 설정 주의*/
for (int j = Lable; j < New_Lable; ++j)
{
Delta[j] += (Sigmoid[j] * (1 - Sigmoid[j]) * Weight_Output_Layer[carry] * Delta_Output[i]);
++carry;
}
}
inc = 0;
carry = 0;
/*Hidden Layer들 사이의 Delta*/
for (int i = Number_Layer - 1; i > 0; --i)
{
carry += Hd_L_Number;
//ex) 30 - (10+jump) < 25 - jump -> 1. 20 < 25 2. 15 < 20 3.10 < 15
for (int z = New_Lable - (2 * Hd_L_Number + jump); z < New_Lable - Hd_L_Number - jump; z++)
{
//ex) 30 - carry < 30 - jump 1. 25 < 30 2. 20 < 25 ...
for (int j = (New_Lable - carry); j < New_Lable - jump; j++)
{
Delta[z] += (Sigmoid[z] * (1 - Sigmoid[z])) * Delta[j] * Weight_Layer[i - 1, inc + small_jump];
small_jump += Hd_L_Number;
}
small_jump = 0;
jump += Hd_L_Number;
inc++;
}
}
carry = 0;
inc = 0;
jump = 0;
/*Weight 갱신*/
//Bias 부분
for (int i = 0; i < New_Lable; ++i)
{
Bias_Weight[i] = (L_N_G * Delta[i] * Bias) + Bias_Weight[i];
}
for (int i = New_Lable; i < Number_Neurons; ++i)
{
Bias_Weight[i] = (L_N_G * Delta_Output[i - New_Lable] * Bias) + Bias_Weight[i];
}
//Input <---> Hidden Layer 1층 부분
//ex) 5 * 5 -> 25
for (int i = 0; i < (Input_Neuron * Hd_L_Number); ++i)
{
carry = i % Input_Neuron; //Input 2개 일때 (l--> 0 1 0 1)
if (i > 0)
{ //i--> 0 1 2 3 4 5 6 7 => l --> 0 1 0 1 0 1 0 1
if (carry == 0)
{
++k; //K--> 0 0 1 1 2 2 3 3
}
}
Weight_Input_Layer[i] = (L_N_G * Delta[k] * Input[carry + bnc * Input_Neuron]) + Weight_Input_Layer[i];
}
carry = 0;
k = 0;
/*Hidden Layer 사이의 Weight 갱신*/
for (int i = (Number_Layer - 1); i > 0; --i)
{
carry += Hd_L_Number;
//ex) 1. 25 - 5 - 5 < 25 - 5 2. 25 - 10 -5 < 25 - 10
for (int j = (New_Lable - carry - Hd_L_Number); j < (New_Lable - carry); ++j)
{
//ex) 1. 25 - 5 < 25 - 0 2. 25-10 < 25-5 ...
for (int k = (New_Lable - carry); k < (New_Lable - jump); ++k)
{
Weight_Layer[i - 1, inc] = (L_N_G * Delta[k] * Sigmoid[j]) + Weight_Layer[i - 1, inc];
++inc;
}
}
jump += Hidden_Layer[i];
inc = 0;
}
inc = 0;
jump = 0;
carry = 0;
//Hidden Layer(마지막 층) <---> Output Layer
//ex) 1. 0 < 5*5
for (int i = 0; i < (Output_Neuron * Hd_L_Number); ++i)
{
carry = i % Hd_L_Number; // 0 1 2 3 4 0 1 2 3 4
if (i > 0 && ((i % Hd_L_Number) == 0))
{
++k; //ex) i: 5->0 10->0 15->0 20->0
}
Weight_Output_Layer[i] = (L_N_G * Delta_Output[k] * Sigmoid[carry + Lable]) + Weight_Output_Layer[i];
}
carry = 0;
k = 0;
/*Delta += 사용하였기 때문에 초기화 해주어야 함*/
for (int i = 0; i < New_Lable; ++i)
{
Delta[i] = 0;
}
for (int i = 0; i < Output_Neuron; ++i)
{
Delta_Output[i] = 0;
}
/*Sum += 문법 초기화*/
for (int i = 0; i < New_Lable; ++i)
{
Sum[i] = 0;
}
for (int i = 0; i < Output_Neuron; ++i)
{
Sum_Output[i] = 0;
}
/*최종 Error 값 구하는 곳*/
//Mean Square Error 적용해보기
//RMSE = Root *( (1.0 / n) * Sigma(i) * pow((Target_Vector(i) - Output(i)) , 2) )
//루트 --> sqrt(실수) , 제곱 --> pow(a , 2)
for (int i = 0; i < Output_Neuron; ++i)
{
Error[i] = (Target_t[bnc, i] - Sigmoid_Output[i]);
RMSE += ((1.0 / Output_Neuron) * Math.Pow(Error[i], 2));
Error_add[i] += Math.Abs(Error[i]);
}
RMSE = Math.Sqrt(RMSE);
++bnc;
++Iteration;
if (bnc == Get_days)
{
bnc = 0;
}
if ((Iteration % Get_days) == 0)
{
RMSE = 0;
for (int i = 0; i < Output_Neuron; ++i)
{
Error_Result[i] = (Error_add[i] / Get_days);
Error_add[i] = 0;
}
}
}//while(Epoch-- > 0)
//TEST OUTPUT(TEXT Result)
/*Test할 Input 값 입력*/
bnc = 0;
#if false
for (int i = 0; i < Input_Neuron; i++)
{
for (int j = 0; j < Get_days; j++)
{
//ex) 0,4,8,12,16 || 0 4 8 12 16
//ex) 1,5,9,13,17 || 1 5 9 13
T_Input[i] += Input[j * Input_Neuron + i];
}
T_Input[i] = T_Input[i] / Get_days;
}
#endif
//1.시가 2.고가 3.저가 4.거래량 5.종가(타겟)
T_Input[0] = stock_bp[0].s_dmp_int;
T_Input[1] = stock_bp[0].s_dhp_int;
T_Input[2] = stock_bp[0].s_dlp_int;
T_Input[3] = stock_bp[0].s_dtv_int;
T_Input[0] /= Math.Pow(10, digits);
T_Input[1] /= Math.Pow(10, digits);
T_Input[2] /= Math.Pow(10, digits);
T_Input[3] /= Math.Pow(10, digits_tv);
/*Input - Hidden Layer[0] 사이 Sum,Sigmoid,Delta */
for (int i = 0; i < Hd_L_Number; ++i)
{
for (int j = 0; j < Input_Neuron; ++j)
{
T_Sum[i] += (T_Input[j + bnc * 2] * Weight_Input_Layer[inc]);
++inc;
}
T_Sum[i] += (Bias * Bias_Weight[i]);
T_Sigmoid[i] = (1.0 / (1.0 + Math.Exp(-T_Sum[i])));
}
inc = 0;
/*Hidden Layer들 사이의 Sum, Sigmoid*/
for (int i = 0; i < (Number_Layer - 1); ++i)
{
carry += Hd_L_Number;
for (int j = carry; j < carry + Hd_L_Number; ++j)
{
for (int k = jump; k < carry; ++k)
{
T_Sum[j] += (T_Sigmoid[k] * Weight_Layer[i, inc]);
++inc;
}
T_Sum[j] += (Bias * Bias_Weight[j]);
T_Sigmoid[j] = (1.0 / (1.0 + Math.Exp(-T_Sum[j])));
}
inc = 0;
jump += Hd_L_Number;
}
jump = 0;
carry = 0;
/* Output Layer와 연결된 Hidden Layer이용하여 Output Sum,Sigmoid */
for (int i = 0; i < Output_Neuron; ++i)
{
for (int j = Lable; j < New_Lable; ++j)
{
T_Output_Sum[i] += (T_Sigmoid[j] * Weight_Output_Layer[inc]);
++inc;
}
T_Output_Sum[i] += (Bias * Bias_Weight[New_Lable + i]);
T_Output_Sigmoid[i] = (1.0 / (1.0 + Math.Exp(-T_Output_Sum[i])));
}
inc = 0;
return(Math.Abs(T_Output_Sigmoid[0]) * Math.Pow(10, digits));
} //BP_START
public stock_ html_get_event(stock_ get)
{
return(get);
}
