Neural network fitting tool for multiple experimental data

Question

Carlo Campigotto 2021-6-8

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/850965-neural-network-fitting-tool-for-multiple-experimental-data

回答： KALASH 2024-5-3

Hi,

I'd like to use neural networks to fit data from multiple experiments, let's say 10 experiments. During each experiment, boundary conditions (my inputs) are constant, but differ from one experiment to another. The output is a time-temperature matrix. I was thinking I could use something like the built-in "body fat percentage" example, but then I should use a matrix for every experiment instead of a column for every person. Is it possible?

Thank you

ps: body fat example

bodyfatInputs - a 13x252 matrix defining thirteen attributes for 252 people.

bodyfatTargets - a 1x252 matrix of associated body fat percentages, to be estimated from the inputs.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

KALASH 2024-5-3

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/850965-neural-network-fitting-tool-for-multiple-experimental-data#answer_1451851

Hi Carlo,

You can try flattening or reshaping each time-temp matrix into a vector so that each experiment’s output can be a column in your target matrix (the body fat percentage which you have given). This assumes the time-temp matrices are same for each experiment.

For eg : if your time-temp matrix for one experiment is 10*10, then you would reshape it into a 100*1 vector.

For preparing your inputs, ensure that each set of boundary conditions is represented as a column in your input matrix and for multiple boundary conditions, each should be a row in the matrix, with each column representing a different experiment.

You can then continue with your construction of the neural network as shown below and train it.

You can refer to the below code snippet:

% Assuming you have 5 boundary conditions and 100 time-temperature data points per experiment

% For 10 experiments

numBoundaryConditions = 5;

numDataPoints = 100; % For a 10x10 time-temperature matrix flattened

numExperiments = 10;

% Randomly generated example data

inputs = rand(numBoundaryConditions, numExperiments); % Replace with your actual data

targets = rand(numDataPoints, numExperiments); % Replace with your actual flattened matrices

% Create a Feedforward Neural Network

net = feedforwardnet([10 10]); % Example architecture, adjust as needed