Predictor
This source code was originally to predict lightning strikes. This source code is run in MATLAB and can be modified for different no of inputs. Meaning, this source code can be used for any kind of prediction!
Original Author: Sara Khoo
Inputs
The code is in two parts. The training and recall phase (see the source codes for both are given) make up the whole prediction. The training requires a series of input values of the parameter intended for prediction. The recall phase requires the saved weights from the training phase.
Assumptions
The user needs a basic understanding of Neural Networks and Back Propagation Theory.
Returns
The code returns a nice prediction graphically plotted with MATLAB.
Code
%Hidden Layer - Gaussian Activation Function
%Input and Output Layers - Linear
%MaxDataNo = No. of Years x 24
%Open and read file
disp(' ');
file=input('Please enter path of data file:', 's');
fid=fopen(file,'rt');
[DataArray]=fscanf(fid,'%d');
x=max(DataArray);
DataArray=DataArray/x;
DataArray2=DataArray;
%Initialization
InputUnitNo=24;
HiddenUnitNo=10;
OutputUnitNo=1;
MaxDataNo=120;
MaxPatternNo=MaxDataNo-InputUnitNo;
Eta=0.2;
Alpha=0.2;
Error=0.13;
Epoch=0;
RMSError=1;
%Weights
bias_k=rand(OutputUnitNo,1);
dbias_k=0;
bias_j=rand(HiddenUnitNo,1);
dbias_j=0;
vji=rand(HiddenUnitNo,InputUnitNo);
dvji=0;
wkj=rand(OutputUnitNo,HiddenUnitNo);
dwkj=0;
while RMSError>Error,
if Epoch>1000,
break;
else
for i=1:MaxPatternNo,
%Start to learn
Input=DataArray2(i:i+23);
Target=DataArray(i+24);
%Count netj
netj=bias_j+vji*Input;
%Gaussian
HiddenOutput=exp(-(netj.^2));
%Count netk
netk=bias_k+wkj*HiddenOutput;
Output=netk;
%Correction Term
delk=(Target-Output);
delj=((-2*(netj.')*HiddenOutput)*(delk*wkj).');
%Update Weights
dwkj=Eta*delk*(HiddenOutput.')+Alpha*dwkj;
dvji=Eta*delj*(Input.')+Alpha*dvji;
wkj=wkj+dwkj;
vji=vji+dvji;
%Update Biases
dbias_k=Eta*delk+Alpha*dbias_k;
dbias_j=Eta*delj+Alpha*dbias_j;
bias_k=bias_k+dbias_k;
bias_j=bias_j+dbias_j;
%Calculating Error Function
ErrorFunc(i)=((Target-Output).^2);
%Update Data Array with Output Value
DataArray2(i+24)=Output;
end
RMSError=0.5*sum(ErrorFunc)
Epoch=Epoch+1
end
end
%Close Data File
fclose(fid);
disp('Weights are saved in C:My Documents by default');
%Save Weights
fid2=fopen('C:My DocumentsWkj.txt','wt');
fprintf(fid2,'%1.4f
',wkj);
fclose(fid2);
fid3=fopen('C:My DocumentsVji.txt','wt');
fprintf(fid3,'%1.4f
',vji);
fclose(fid3);
%Save Biases
fid4=fopen('C:My DocumentsBiask.txt','wt');
fprintf(fid4,'%1.4f
',bias_k);
fclose(fid4);
fid5=fopen('C:My DocumentsBiasj.txt','wt');
fprintf(fid5,'%1.4f
',bias_j);
fclose(fid5);
DataArray=x*DataArray;
DataArray2=x*DataArray2;
Error=abs(DataArray-DataArray2);
i=0:MaxDataNo-1;
subplot(2,1,1);
hold on
plot(i,DataArray2(1:MaxDataNo),'m-.');
plot(i,DataArray(1:MaxDataNo));
subplot(2,1,2);
plot(i,Error(1:MaxDataNo),'g');
==================
%Recall Phase
%Load Data,Weights and Biases
disp(' ');
file=input('Please enter path of data file:', 's');
fid=fopen(file,'rt');
[DataArray]=fscanf(fid, '%d');
x=max(DataArray);
DataArray=DataArray/x;
fclose(fid);
disp(' ');
file=input('Please enter path of wkj file:', 's');
fid=fopen(file,'rt');
[readwkj]=fscanf(fid,'%f')
fclose(fid);
disp(' ');
file=input('Please enter path of vji file:', 's');
fid=fopen(file,'rt');
[readvji]=fscanf(fid,'%f');
fclose(fid);
disp(' ');
file=input('Please enter path of bias_k file:', 's');
fid=fopen(file,'rt');
[readbias_k]=fscanf(fid,'%f');
fclose(fid);
disp(' ');
file=input('Please enter in path of bias_j file:', 's');
fid=fopen(file,'rt');
[readbias_j]=fscanf(fid,'%f');
fclose(fid);
%Initialization
InputUnitNo=24;
HiddenUnitNo=10;
OutputUnitNo=1;
MaxDataNo=144;
MaxPatternNo=MaxDataNo-InputUnitNo;
PredictionDataNo=24;
DataArray2=[DataArray(1:MaxDataNo-PredictionDataNo);zeros(PredictionDataNo,1)];
%Initialize biases and weights
bias_j=readbias_j;
wkj=readwkj.';
bias_k=readbias_k;
%Convert vji to matrix
vji=readvji(1:HiddenUnitNo);
for i=11:10:InputUnitNo*HiddenUnitNo-9,
vji=[vji,readvji(i:i+9)];
end
for i=1:MaxPatternNo,
Input=DataArray2(i:i+23);
%Count netj
netj=bias_j+vji*Input;
%Gaussian
HiddenOutput=exp(-(netj.^2));
%Count netk
netk=bias_k+wkj*HiddenOutput;
Output=abs(netk);
%Update Data Array with Output Value
DataArray2(i+24)=Output;
end
DataArray2=DataArray2*x;
DataArray=DataArray*x;
i=0:MaxDataNo-1;
hold on
plot(i,DataArray2,'m-.');
plot(i,DataArray);
Error=abs(DataArray-DataArray2);
i=0:MaxDataNo-1;
subplot(2,1,1);
hold on
plot(i,DataArray2(1:MaxDataNo),'m-.');
plot(i,DataArray(1:MaxDataNo));
subplot(2,1,2);
plot(i,Error(1:MaxDataNo),'g');
title('Recall Phase using weights with RMS Error = 0.1163')
Loading Comments ...
Comments
No comments have been added for this post.
You must be logged in to make a comment.