Multi parameter input multiparameter (x,y graph) output model for prediction

Good afternoon,

I have a dataset containing 1000 different scenarios of various materials with varying material properties and loading; the data also contains the expected number of cycles (life). My goal is to create a map between material properties and their respective wholer curve e.g (load vs cycles to failure). For further background the data was constructed using a latin hypercube sampling methodology.

The ideal being something like

(val prop1,valprop2...valpropn) -> Load, Life

5 5000

10 3000

16 1000

The inverse of this is also acceptable e.g

Load, Life -> (val prop1,valprop2...valpropn)

5 5000

10 3000

16 1000

My problem is Load and Life are linked and can't be seperated. I have thought of this problem in a few different ways, a straight regression problem which didn't perform particularly well. I have also been trying a sequence to sequence approach however, most literature in matlab points me down the LSTM route which often focuses on sequence to one or; sequence to sequence with a single feature. I have some sample code below that I have been trying but I get a strange error that appears incomplete. Any thoughts on how to do this would be greatly appreciated.

In the below example: Material Parameters is 1000 samples of a single row containing 10 materail propeties as features

In this trial Load Life is intended as an XY plot where the first 5 enteries are x values and the next 5 are y values. Not sure if this is best practice or not I was just trying to get soemthing to work....

MaterialParameters=reshape(MaterialParameters, [10, 1, 1000]);

LoadLife=reshape(LoadLife, [1, 10, 1400]);

% Define input and output sequences (non-time series)

inputSequence = MaterialParameters; % 10 features, 1 channel, 100 sequences

outputSequence = LoadLife; % 5 features, 1 channel, 100 sequences

% Define the layers of the sequence-to-sequence network

layers = [

sequenceInputLayer(10) % Input layer for 10 features

lstmLayer(50, 'OutputMode', 'sequence') % LSTM layer with 50 hidden units

lstmLayer(50, 'OutputMode', 'sequence') % Second LSTM layer

fullyConnectedLayer(10) % Fully connected layer for 10 output features

regressionLayer]; % Regression layer for continuous output

% Specify training options

options = trainingOptions('adam', ...

'MaxEpochs', 100, ...

'MiniBatchSize', 20, ...

'Verbose', 1, ...

'Plots', 'training-progress');

% Train the network

net = trainNetwork(inputSequence, outputSequence, layers, options);

Error:

Error using trainNetwork

The training sequences are of feature dimension 10 1 but the input layer expects sequences of feature dimension 10.

Error in MaterialExtractor (line 325)

net = trainNetwork(inputSequence, outputSequence, layers, options);

Any help would be greatly appreciated!

Answer 1

[Image Analyst]

Sounds like a partial least squares situation since variables are related. However, why don't you give the "Regression Learner" app a try. It's on the Apps tab of the tool ribbon.

[alex bradley]

I have tried this actually and it appears that the regression learner app can only do multi input single output analyses. For me I think I need to be able to output the whole array