optimal values for cell data

By following answer here, i tried to optimize machine learning parameters using baysopt. However, i am getting error :

I

[Walter Roberson]

We are getting pretty strict there needing to be a really good reason for deleting a Question that has a valid attempt at an Answer. We are having too much trouble with people attempting to use the resource for free private consulting -- posting something, getting their free answer, and then demanding that the question be removed.

[Stephen23]

@Ali: when you posted your question here you agreed to the terms and conditions given here:

https://www.mathworks.com/matlabcentral/termsofuse.html#answers

This means that your posted content is released under Creative Commons Attribution Share Alike 3.0 license, and as such anyone can copy and distribute your posted content with the appropriate attribution.

Internet archiving websites have likely already saved publicly-available copies of this page.

Summary: if you do not want sensitive data available on the internet, do not post it.

[Walter Roberson]

• Every student claims that their code is "sensitive" on the grounds that other students might read the posting and copy from them.
• There are no National Security considerations here.
• If there are Trade Secret matters here, then legally speaking you destroyed the "secret" as soon as you posted the material (Trade Secret case law is really strict on that point. Like if a piece of paper with a Trade Secret blows out of your hand and someone finds it, then you have just lost Trade Secret status.)
• You would have a difficult time convincing us that you are working on a Patent: people working on Patents know to hire consultants with Non-Disclosure Agreements

When I look at your previous postings, the most generous reading I can come up with is that you must might be working on a thesis. For thesis, the important part is that the ideas are yours; it is permitted to seek assistance with implementation .

[Rik]

Text of the flag by Ali:

I want to hide or delete this question as it contains sensitive material (code) of my project. I am grateful to MATLAB that the issue has been resolved.

Answer 1

[Nipun Katyal]

Yeah there is some problem while handling the cell

Here is the correct way to handle it

% Make some data
Daten = rand(100, 3);
Daten(:,3) = Daten(:,1) + Daten(:,2) + .1*randn(100, 1);  % Minimum asymptotic error is .1
[m,n] = size(Daten) ;
% Split into train and test
P = 0.7 ;
Training = Daten(1:round(P*m),:) ; 
Testing = Daten(round(P*m)+1:end,:);

XTr = Training(:,1:n-1); 
YTr = Training(:,n);
XTe = Testing(:,1:n-1);
YTe = Testing(:,n);
XTrain=num2cell(XTr(:,1));
YTrain=num2cell(YTr(:,1));
XTest=num2cell(XTe);
YTest=num2cell(YTe);
% Define a train/validation split to use inside the objective function
cv = cvpartition(numel(YTrain), 'Holdout', 1/3);
% Define hyperparameters to optimize
vars = [optimizableVariable('hiddenLayerSize', [1,20], 'Type', 'integer');
	    optimizableVariable('lr', [1e-3 1], 'Transform', 'log')];
% Optimize
minfn = @(T)kfoldLoss(XTrain', YTrain', cv, T.hiddenLayerSize, T.lr);
results = bayesopt(minfn, vars,'IsObjectiveDeterministic', false,...
    'AcquisitionFunctionName', 'expected-improvement-plus');
T = bestPoint(results);
function rmse = kfoldLoss(x, y, cv, numHid, lr)
% Train net.
net = feedforwardnet(numHid, 'traingd');
net.trainParam.lr = lr;
net = train(net, x(:,cv.training), y(:,cv.training));
% Evaluate on validation set and compute rmse
ypred = net(x(:, cv.test));
n = size(ypred);
pw = 2*ones(n);
pw = num2cell(pw);
cMinus = cellfun(@minus, ypred, y(cv.test), 'UniformOutput', false);
cSquare = cellfun(@power, cMinus, pw, 'UniformOutput', false);
cSquareVect = cell2mat(cSquare);
cMean = mean(cSquareVect);
rmse = sqrt(cMean);
%rmse = sqrt(mean((ypred - y(cv.test)).^2));
end

[Ali]

Hi Nipun!! Thank you once again but i am still getting some errors as shown in attached file.

[Nipun Katyal]

The initial learning rate is high which leads to quickly diverging values and causes the net function to return Nan values and hence the objective is also Nan,

You can try reducing the learning rate for example 1e-3 to 1e-1,

% Make some data
Daten = rand(100, 3);
Daten(:,3) = Daten(:,1) + Daten(:,2) + .1*randn(100, 1);  % Minimum asymptotic error is .1
[m,n] = size(Daten) ;
% Split into train and test
P = 0.7 ;
Training = Daten(1:round(P*m),:) ; 
Testing = Daten(round(P*m)+1:end,:);
XTr = Training(:,1:n-1); 
YTr = Training(:,n);
XTe = Testing(:,1:n-1);
YTe = Testing(:,n);
XTrain=num2cell(XTr(:,1));
YTrain=num2cell(YTr(:,1));
XTest=num2cell(XTe);
YTest=num2cell(YTe);
% Define a train/validation split to use inside the objective function
cv = cvpartition(numel(YTrain), 'Holdout', 1/3);
% Define hyperparameters to optimize
vars = [optimizableVariable('hiddenLayerSize', [1,20], 'Type', 'integer');
	    optimizableVariable('lr', [1e-3 1e-1], 'Transform', 'log')];
% Optimize
minfn = @(T)kfoldLoss(XTrain', YTrain', cv, T.hiddenLayerSize, T.lr);
results = bayesopt(minfn, vars,'IsObjectiveDeterministic', false,...
    'AcquisitionFunctionName', 'expected-improvement-plus');
T = bestPoint(results);
function rmse = kfoldLoss(x, y, cv, numHid, lr)
% Train net.
net = feedforwardnet(numHid, 'traingd');
net.trainParam.lr = lr;
net = train(net, x(:,cv.training), y(:,cv.training));
% Evaluate on validation set and compute rmse
ypred = net(x(:, cv.test));
n = size(ypred);
pw = 2*ones(n);
pw = num2cell(pw);
cMinus = cellfun(@minus, ypred, y(cv.test), 'UniformOutput', false);
cSquare = cellfun(@power, cMinus, pw, 'UniformOutput', false);
cSquareVect = cell2mat(cSquare);
cMean = mean(cSquareVect);
rmse = sqrt(cMean);
%rmse = sqrt(mean((ypred - y(cv.test)).^2));
end

[Nipun Katyal]

This should do

% Make some data
Daten = rand(100, 3);
Daten(:,3) = Daten(:,1) + Daten(:,2) + .1*randn(100, 1);  % Minimum asymptotic error is .1
[m,n] = size(Daten) ;
% Split into train and test
P = 0.7 ;
Training = Daten(1:round(P*m),:) ; 
Testing = Daten(round(P*m)+1:end,:);
XTr = Training(:,1:n-1); 
YTr = Training(:,n);
XTe = Testing(:,1:n-1);
YTe = Testing(:,n);
XTrain=num2cell(XTr(:,1));
YTrain=num2cell(YTr(:,1));
XTest=num2cell(XTe);
YTest=num2cell(YTe);
% Define a train/validation split to use inside the objective function
cv = cvpartition(numel(YTrain), 'Holdout', 1/3);
% Define hyperparameters to optimize
vars = [optimizableVariable('hiddenLayerSize', [1,20], 'Type', 'integer');
	    optimizableVariable('lr', [1e-3 1e-1], 'Transform', 'log')];
% Optimize
minfn = @(T)kfoldLoss(XTrain', YTrain', cv, T.hiddenLayerSize, T.lr);
results = bayesopt(minfn, vars,'IsObjectiveDeterministic', false,...
    'AcquisitionFunctionName', 'expected-improvement-plus');
T = bestPoint(results); 
% Train final model on full training set using the best hyperparameters
net = layrecnet(1:2,T.hiddenLayerSize, 'traingd');
net.trainParam.lr = T.lr;
net = train(net, XTrain', YTrain');
% Evaluate on test set and compute final rmse
% ypred = net(XTest');
% finalrmse = sqrt(mean((ypred - YTest').^2))
% Evaluate on validation set and compute rmse
ypred = net(XTest(:,1)');
n = size(ypred);
pw = 2*ones(n);
pw = num2cell(pw);
cMinus = cellfun(@minus, ypred, YTest', 'UniformOutput', false);
cSquare = cellfun(@power, cMinus, pw, 'UniformOutput', false);
cSquareVect = cell2mat(cSquare);
cMean = mean(cSquareVect);
Rmse = sqrt(cMean)

function rmse = kfoldLoss(x, y, cv, numHid, lr)
% Train net.
net = feedforwardnet(numHid, 'traingd');
net.trainParam.lr = lr;
net = train(net, x(:,cv.training), y(:,cv.training));
% Evaluate on validation set and compute rmse
ypred = net(x(:, cv.test));
n = size(ypred);
pw = 2*ones(n);
pw = num2cell(pw);
cMinus = cellfun(@minus, ypred, y(cv.test), 'UniformOutput', false);
cSquare = cellfun(@power, cMinus, pw, 'UniformOutput', false);
cSquareVect = cell2mat(cSquare);
cMean = mean(cSquareVect);
rmse = sqrt(cMean);
%rmse = sqrt(mean((ypred - y(cv.test)).^2));
end

[Ali]

Thank you so much Nipun!!

I want to delete this question from MATLAB sentral as it contains some sensitive information ? Is it possible as been discussed here.

Answer 2

[Nipun Katyal]

Inorder to perform operations on cells use cellfun as mentioned below:

% Make some data
Daten = rand(100, 3);
Daten(:,3) = Daten(:,1) + Daten(:,2) + .1*randn(100, 1);  % Minimum asymptotic error is .1
[m,n] = size(Daten) ;
% Split into train and test
P = 0.7 ;
Training = Daten(1:round(P*m),:) ; 
Testing = Daten(round(P*m)+1:end,:);

XTr = Training(:,1:n-1); 
YTr = Training(:,n);
XTe = Testing(:,1:n-1);
YTe = Testing(:,n);
XTrain=num2cell(XTr(:,1));
YTrain=num2cell(YTr(:,1));
XTest=num2cell(XTe);
YTest=num2cell(YTe);
% Define a train/validation split to use inside the objective function
cv = cvpartition(numel(YTrain), 'Holdout', 1/3);
% Define hyperparameters to optimize
vars = [optimizableVariable('hiddenLayerSize', [1,20], 'Type', 'integer');
	    optimizableVariable('lr', [1e-3 1], 'Transform', 'log')];
% Optimize
minfn = @(T)kfoldLoss(XTrain', YTrain', cv, T.hiddenLayerSize, T.lr);
results = bayesopt(minfn, vars,'IsObjectiveDeterministic', false,...
    'AcquisitionFunctionName', 'expected-improvement-plus');
T = bestPoint(results);
function rmse = kfoldLoss(x, y, cv, numHid, lr)
% Train net.
net = feedforwardnet(numHid, 'traingd');
net.trainParam.lr = lr;
net = train(net, x(:,cv.training), y(:,cv.training));
% Evaluate on validation set and compute rmse
ypred = net(x(:, cv.test));
cMinus = cellfun(@minus, ypred, y(cv.test), 'UniformOutput', false);
cMean = cellfun(@mean, cMinus);
rmse = sqrt(cMean);
%rmse = sqrt(mean((ypred - y(cv.test)).^2));
end

[Nipun Katyal]

In your case the rmse function will be:

function rmse = kfoldLoss(x, y, cv, numHid, lr)
% Train net.
net = feedforwardnet(numHid, 'traingd');
net.trainParam.lr = lr;
net = train(net, x(:,cv.training), y(:,cv.training));
% Evaluate on validation set and compute rmse
ypred = net(x(:, cv.test));
n = size(ypred);
pw = 2*ones(n);
pw = num2cell(pw);
cMinus = cellfun(@minus, ypred, y(cv.test), 'UniformOutput', false);
cSquare = cellfun(@power, cMinus, pw, 'UniformOutput', false);
cMean = cellfun(@mean, cSquare);
rmse = sqrt(cMean);
%rmse = sqrt(mean((ypred - y(cv.test)).^2));
end