CreateNeuralNetworkinExcel

Arryex,2012
1. PreparetrainingdataforNN
a. Identifywhatareinputsandoutput
b. Howmanyinputsandoutput?
c. Howmanypatterns?
d. Performdatascalinginto[0,1],[1,1],etc
2. DefineNNConfiguration
a. HowmanySlabs?
b. HowmanyHiddenneuronsoneachslab?
c. Definetransferfunction,exampletanh=1/exp(x))
d. Initiateweightandbiasvalues
e. Insertformulaforeachhiddenneurons
f. Insertformulaforoutputneurons
g. Insertformulaforerror^2andMSE
3. RunSolveraddin
a. CriteriacanbeminimizeMSEormaximizeR2
b. InsertGraphforoutputandpredictedoutput
4. Postprocessingdata
a. Revertscaleofpredictedoutput
b. Recordfinalvaluesofweightandbias
5. DeployTrainedNet
a. ConvertformularelationbetweeninputandoutputintoCorMQ4code
b. CreatedllfileusingCcompilerorMT4
c. CreateMT4indicatorbasedontrainednet.

Example
1. Wehavetimeseriesdata(MackeyGlass)thatconsistof4inputsand1input,700patterns
2. Scaleinputandoutputdatabecome[0,1]

InputOutputandScaledData
3. DefineNNconfiguration
a. NNisbackpropagation,3slabs
b. 4inputsinSlab1
c. 29hiddenneuronsinSlab2
d. Thereare29weightforeachinput,29bias,thenwehave29x4weighttoconnectSlab
1toSlab2
e. Thereare29weightandonebiastoconnectSlab2toSlab3


NetConfiguration

NetworkConfiguration
Input:X,X1,X2,X3
Slab2Calculation:
Weightdata:
w(1,1),w(1,2),w(1,3),w(1,4)forhiddenneuron(1,1)
w(2,1),w(2,2),w(2,3),w(2,4)forhiddenneuron(1,2)
..
w(1,1),w(1,2),w(1,3),w(1,4)forhiddenneuron(1,29)
Bias:
b(1,1)forhiddenneuron(1,1)
b(1,2)forhiddenneuron(1,2)
.
b(1,29)forhiddenneuron(1,9)

ExcelFormula:
HiddenNeuron(1,1)=1/(1+(EXP((X*w(1,1)+X1*w(1,2)+X2*w(1,3)+X3*w(1,4)+b(1,1))
HiddenNeuron(1,2)=1/(1+(EXP((X*w(2,1)+X1*w(2,2)+X2*w(2,3)+X3*w(2,4)+b(1,2))
.
HiddenNeuron(1,29)=1/(1+(EXP((X*w(29,1)+X1*w(29,2)+X2*w(29,3)+X3*w(29,4)+b(1,29))

Slab3Calculation
Weightdata:
w(2,1),w(2,2),w(2,3),w(2,4),w(2,5).,w(2,29)
Bias:
b(2,1)
ExcelFormula:
OutputNeuron=1/(1+(EXP((h(1,1)*w(2,1)+h(1,2)*w(1,2)+h(1,3)*w(1,3)+.h(1,29)*w(29,4)+b(2,1))

CalculateError^2,MSE
Formula:
Error^2=(outputpredictionX4)^2
MSE=Average(Error^2)
RunSolveraddin

AdjustableCells:allweightandbiasdata
Criteria:MinimizeMSE
DisplayChartpredictedanddesiredoutput