Error in using LSTM for multipe responses

clear all; clc;

Data=xlsread("C:\Users\vsvbh\OneDrive\Documents\LSTMPRBS.xlsx")

height1=Data(:,2)

height2=Data(:,3)

height3=Data(:,4)

flowrate1=Data(:,5)

flowrate2=Data(:,6)

[row col]=size(Data);

%create database for lstm

SequenceLength=1

Database={}

resultDatabase=[]

count=0

for i=row:-1:SequenceLength

count=count+1

a=height1(i-SequenceLength+1:i)';

a=flip(a);

b=height2(i-SequenceLength+1:i)';

b=flip(b);

c=height3(i-SequenceLength+1:i)';

c=flip(c);

d=flowrate1(i-SequenceLength+1:i)';

d=flip(d);

e=flowrate2(i-SequenceLength+1:i)';

e=flip(e);

Database{count,1}=[a;b;c;d;e];

end

[rt ct]=size(Database)

count=0

for i=1:rt-1

count=count+1

a=Database{count,1};

b=Database{count,1};

c=Database{count,1};

resultDatabase1(count,1)=a(1,end);

resultDatabase2(count,1)=b(2,end);

resultDatabase3(count,1)=c(3,end);

end

HeightDatabase=Database;

save HeightDatabase

%lstm creation

for i=1:rt-300

stockTrain{i,1}=Database{i,1}

end

restTrain1=resultDatabase1(1:rt-300)

restTrain2=resultDatabase2(1:rt-300)

restTrain3=resultDatabase3(1:rt-300)

restTrain={}

for i=1:700

restTrain{i,1}={restTrain1(i,:);restTrain2(i,:);restTrain3(i,:)}

end

numFeatures=5

numHiddenLayers1=100

numHiddenLayers2=100

numResponses=3

Layers=[ sequenceInputLayer(numFeatures),...

lstmLayer(numHiddenLayers1,'OutputMode','Last'),...

lstmLayer(numHiddenLayers2,'OutputMode','Last'),...

fullyConnectedLayer(numResponses),...

regressionLayer

]

miniBatchSize=20;

Epoch=200;

options=trainingOptions('adam',...

'ExecutionEnvironment','auto',...

'MaxEpochs',Epoch,...

'MiniBatchSize',miniBatchSize,...

'plots','training-progress')

Trainedheight=trainNetwork(stockTrain,restTrain,Layers,options)

save Trainedheight

i need to trian the data for three responses i have 1000 data points out of which i used 700 to train . so i need to train the data for all the three but it was giving a error

Answer 1

[Milan Bansal]

Hi,

In my understanding you are facing an error when training regression LSTM (Long Short Term Memory) model for multiple responses (3 in this case).

The issue is with the data type of "restTrain" which is a "cell array". The target response in "trainNetwork" function for the regression task must be a vector or a matrix.

Create "restTrain" as a matrix instead of a cell array. See the code below.

restTrain = [restTrain1 restTrain2 restTrain3]; 

Refer to the documentation link to know more about format of target responses in "trainNetwork" functionf for different tasks.

https://www.mathworks.com/help/deeplearning/ref/trainnetwork.html#mw_d0b3a2e4-09a0-42f9-a273-2bb25956fe66