Bayesopt not printing result

Hi. Im trying to optimize hyperparameters in my LSTM. Im am using Bayesopt. However bayesopt did not print any result for the optimize hyperparameters (I've even put Verbose = 1 . Here is my code

% SPECIFY TRAIN DATA FILE PATH

display = 'specify file name? ';

file = input(display);

filePath = fullfile('data',file);

display = sprintf('Loading File Path for %s',file);

disp(display)

clearvars display

% READ TRAIN DATA FILE

dataTrain = readtable(filePath);

display = sprintf('%s read',file);

disp(display)

clearvars display

% DIVIDE DATA INTO TRAIN,VAL,TEST

rows = height(dataTrain)

idx = randperm(rows);

trainIDX = rows*0.8*0.8;

valIDX = rows*0.8*0.2;

testIDX = rows*0.2;

idx1 = (idx<=trainIDX);

idx2 = (idx>trainIDX & idx<=(trainIDX+valIDX));

idx3 = (idx>(trainIDX+valIDX));

trainData = dataTrain(idx1,:);

valData = dataTrain(idx2,:);

testData = dataTrain(idx3,:);

clearvars rows idx trainIDX valIDX testIDX idx1 idx2 idx3

% CALCULATE NO OF TRAIN FLEET

totalFleet = max(trainData{:,1});

display = sprintf('Total No. of train fleet is %d',totalFleet);

disp(display)

clearvars display

% TRAIN DATA NORMALIZATION

minTrain = min(trainData{:,3});

maxTrain = max(trainData{:,3});

subTable1 = normalize(trainData(:,4:12),'range');

subTable2 = trainData(:,1:3);

clearvars trainData

trainData = [subTable2,subTable1];

clearvars subTable1 subTable2

% DIVIDE TRAINING DATA

Xtrain = cell(totalFleet,1);

Ytrain = cell(totalFleet,1);

for i = 1:totalFleet;

M = rows2vars(trainData(trainData.Fleet==i,[4:12]));

M = M(:,2:end);

Xtrain{i,:} = M;

Xtrain{i,:} = table2array(Xtrain{i,:});

clearvars M

M = rows2vars(trainData(trainData.Fleet==i,[3]));

M = M(:,2:end);

Ytrain{i,:} = M;

Ytrain{i,:} = table2array(Ytrain{i,:});

clearvars M totalFleet

end

disp('Training Data Divided into Xtrain and Ytrain')

clearvars i display

% CALCULATE NO OF VAL FLEET

totalFleet = max(valData{:,1});

display = sprintf('Total No. of val fleet is %d',totalFleet);

disp(display)

clearvars display

% VAL DATA NORMALIZATION

subTable1 = normalize(valData(:,4:12),'range');

subTable2 = valData(:,1:3);

clearvars valData

valData = [subTable2,subTable1];

clearvars subTable1 subTable2

% DIVIDE VAL DATA

Xval = cell(totalFleet,1);

Yval = cell(totalFleet,1);

for i = 1:totalFleet;

M = rows2vars(valData(valData.Fleet==i,[4:12]));

M = M(:,2:end);

Xval{i,:} = M;

Xval{i,:} = table2array(Xval{i,:});

clearvars M

M = rows2vars(valData(valData.Fleet==i,[3]));

M = M(:,2:end);

Yval{i,:} = M;

Yval{i,:} = table2array(Yval{i,:});

clearvars M totalFleet

end

disp('Training Data Divided into Xval and Yval')

clearvars i display

% CALCULATE NO OF TEST FLEET

totalTestFleet = max(testData{:,1});

display = sprintf('Total No. of test fleet is %d',totalTestFleet);

disp(display)

clearvars display

% DATA TEST NORMALIZATION

minTest = min(testData{:,3});

maxTest = max(testData{:,3});

subTable1 = normalize(testData(:,4:11),'range');

subTable2 = testData(:,1:3);

clearvars testData

testData = [subTable2,subTable1];

clearvars subTable1 subTable2

% DIVIDE TEST DATA

Xtest = cell(totalTestFleet,1);

Ytest = cell(totalTestFleet,1);

for i = 1:totalTestFleet;

M = rows2vars(testData(testData.Fleet==i,[3:11]));

M = M(:,2:end);

Xtest{i,:} = M;

Xtest{i,:} = table2array(Xtest{i,:});

clearvars M

M = rows2vars(testData(testData.Fleet==i,[3]));

M = M(:,2:end);

Ytest{i,:} = M;

Ytest{i,:} = table2array(Ytest{i,:});

clearvars M

end

disp('Training Data Divided into Xtest and Ytest')

clearvars i display

% BAYESIAN OPTIMISATION

optimVars = [

optimizableVariable('MaxEpochs',[60 1000],'Type','integer')

optimizableVariable('InitialLearnRate',[1e-3 1],'Transform','log')

optimizableVariable('MiniBatchSize',[26 260], 'Type', 'integer')

optimizableVariable('Size',[10 500], 'Type', 'integer')

optimizableVariable('numHiddenUnits',[10 1000],'Type','integer')];

ObjFcn = makeObjFcn(Xtrain,Ytrain,Xval,Yval);

BayesObject = BayesianOptimization(ObjFcn,optimVars, ...

'MaxTime',50400, ...

'IsObjectiveDeterministic',false, ...

'Verbose',1, ...

'UseParallel',false);

bestIdx = BayesObject.IndexOfMinimumTrace(end);

fileName = BayesObject.UserDataTrace{bestIdx};

savedStruct = load(fileName);

valError = savedStruct.valError

% SCRIPT ObjFcn

function ObjFcn = makeObjFcn(Xtrain,Ytrain,Xval,Yval) %#ok<*INUSD>

ObjFcn = @valErrorFun;

function [valError,cons,fileName] = valErrorFun(optimVars) %#ok<STOUT>

layers = [ ...

sequenceInputLayer(9)

lstmLayer(optimVars.numHiddenUnits,'OutputMode','sequence')

fullyConnectedLayer(optimVars.Size)

dropoutLayer(0.5)

fullyConnectedLayer(1)

regressionLayer];

options = trainingOptions('adam', ...

'MaxEpochs',optimVars.MaxEpochs, ...

'ValidationData',{Xval,Yval},...

'ValidationFrequency',10,...

'MiniBatchSize',optimVars.MiniBatchSize, ...

'InitialLearnRate',optimVars.InitialLearnRate, ...

'GradientThreshold',1, ...

'Shuffle','never', ...

'Plots','training-progress',...

'SequenceLength','longest',...

'Verbose',false);

net = trainNetwork(Xtrain,Ytrain,layers,options);

YPredicted = predict(net,Xval);

valError = 1 - mean(YPredicted == Yval);

fileName = num2str(valError) + ".mat";

save(fileName,'net','valError','options')

cons = [];

end

end

Answer 1

[Harsha Priya Daggubati]

Hi,

I think "bayesopt" should be used to perfrom optimzation on Bayesian Classfication. Refer to following links to gain more insight:

https://www.mathworks.com/help/stats/bayesian-optimization-workflow.html#bva8nmd-1

https://www.mathworks.com/help/stats/bayesopt.html

https://www.mathworks.com/help/stats/bayesian-optimization-case-study.html