Compute_CheckLabels.m

Compute_CNNcfg.m

@@ -0,0 +1,70 @@
+function cfg = Compute_CNNcfg(Nsamples)
+
+
+
+% ----------------------------------------
+% 1) Initialize structure
+% 
+cfg                      = [];
+
+
+
+% ----------------------------------------
+% 2) Iteration and epochs: 
+% 
+% Recall that an iteration is one step taken in the gradient descent
+% algorithm towards minimizing the loss function using a mini-batch. In
+% other words, each iteration is an estimation of the gradient and an
+% update of the network parameters
+% 
+cfg.MiniBatchSize        = 128;           % Subset of the training set that is used to evaluate the gradient of the loss function and update the weights 
+cfg.MaxEpochs            = 100;           % An epoch is the full pass of the training algorithm over the entire training set
+cfg.Shuffle              = 'once';        % (never|once|every-epoch) Shuffle the training data before each training epoch, and shuffle the validation data before each network validation
+
+
+
+% ----------------------------------------
+% 3) Validation process
+% 
+% A validation set is used to test learning and generalization during the
+% training process.
+% By default, if the validation loss is larger than or equal to the
+% previously smallest loss five times in a row, then network training
+% stops. To change the number of times that the validation loss is allowed
+% to not decrease before training stops, use the 'ValidationPatience'. You
+% can add additional stopping criteria using output functions
+% 
+cfg.ValidationData       = [];     % ({Xval,Yval}) Used to validate the network at regular intervals during training.
+% cfg.ValidationFrequency  = 20;   % Number of iterations between evaluations of validation metrics. A suggestion is to choose this value so that the network is validated once/twice/.. per epoch.
+cfg.NumValPerEpoch       = 1;      % Number of validations per epoch. This is not a "trainingOptions" parameter but it is used to gently compute the "ValidationFrequency"
+cfg.ValidationFrequency  = floor(Nsamples/cfg.MiniBatchSize/cfg.NumValPerEpoch);
+cfg.ValidationPatience   = 5;      % (scalar|Inf) Turn off the built-in validation stopping criterion (which uses the loss) by setting the 'ValidationPatience' value to Inf.
+cfg.OutputFcn            = '@(info)stopIfAccuracyNotImproving(info,3));'; % The traininf calls the specified functions once before the start of training, after each iteration, and once after training has finished. The training passes a structure containing information in the following fields:
+%cfg.OutputFcn            = '@(info)savetrainingplot(info);';
+
+% ----------------------------------------
+% 4) Learning rate 
+% 
+cfg.InitialLearnRate     = 1.01;        % (default:0.01) If the learning rate is too low, then training takes a long time. If the learning rate is too high, then training might reach a suboptimal result
+cfg.LearnRateSchedule    = 'none';      % (none|piecewise) Option for dropping the learning rate during training. The software updates the learning rate every certain number of epochs by multiplying with a certain factor.
+cfg.LearnRateDropFactor  = 0.1;         % (default:0.1)  Factor for dropping the learning rate. Multiplicative factor to apply to the learning rate every time a certain number of epochs passes. Valid only when the value of LearnRateSchedule is 'piecewise'.
+cfg.LearnRateDropPeriod  = 4;           % (default:10)   Number of epochs for dropping the learning rate. Valid only when the value of LearnRateSchedule is 'piecewise'.
+
+
+
+% ----------------------------------------
+% Otros paramteres
+% 
+% cfg.CheckpointPath       =  % Path for saving checkpoint networks
+% cfg.ExecutionEnvironment =  % (default:??????) (auto|cpu|gpu|multi-gpu|parallel) Hardware resource for training network
+% cfg.L2Regularization     =  % (default:0.0001) Factor for L2 regularizer (weight decay). You can specify a multiplier for the L2 regularizer for network layers with learnable parameters.
+% cfg.Momentum             =  % (default:0.9) Contribution of the gradient step from the previous iteration to the current iteration of the training. A value of 0 means no contribution from the previous step, whereas a value of 1 means maximal contribution from the previous step.
+
+
+
+% ----------------------------------------
+% Progress visualization
+% 
+cfg.Plots                  = 'none';      % (none|training-progress)
+cfg.Verbose                = 1;                        % Indicator to display training progress information in the command window
+cfg.VerboseFrequency       = cfg.ValidationFrequency;  % Frequency of verbose printing, which is the number of iterations between printing to the command window

Compute_CNNconstructXY.m

@@ -0,0 +1,21 @@
+function Data = Compute_CNNconstructXY(Data)
+
+% Convertir Xtfr a X: the data format of X for the CNN should be (16 32 1 Nsamples)
+for i=1:size(Data.Xtfr,1)
+    Data.X(:,:,1,i) = squeeze(Data.Xtfr(i,:,:));
+end
+Data   = rmfield(Data,'Xtfr');
+
+% Save currect Y and create the class label Y vector
+YALL   = Data.Y;
+Data   = rmfield(Data,'Y');
+
+Data.Y      = YALL(:,1);
+Data.YInfo  = YALL(:,2:end);
+% Data.M1     = YALL(:,2);
+% Data.M1     = YALL(:,3);
+% Data.D      = YALL(:,4);
+% Data.SNR    = YALL(:,5);
+
+% Clear garbage
+clear ans doplot IFO Nsamples IndRan IndTrai IndTest Nlabels IndRan i YALL

Compute_CNNeliminateSNR.m

@@ -0,0 +1,61 @@
+function Data = Compute_CNNeliminateSNR(Data,SNR2ELIM)
+
+% SNR2ELIM: SNR limit value to eliminate
+
+% ----------------------------------------
+% Get data for class noise: 1
+Ind_c0              = Data.Y(:,1)==1;
+Y_c0                = Data.Y(Ind_c0,:);
+TFR_c0              = Data.Xtfr(Ind_c0,:,:);
+
+% ----------------------------------------
+% Get data for class gw: 2
+Ind_c1              = Data.Y(:,1)==2;
+Y_c1                = Data.Y(Ind_c1,:);
+TFR_c1              = Data.Xtfr(Ind_c1,:,:);
+
+% ----------------------------------------
+% Eliminate data for which SNR<=10
+Ind2Eli             = Y_c1(:,5)<=SNR2ELIM;
+Y_c0(Ind2Eli,:)     = [];
+Y_c1(Ind2Eli,:)     = [];
+TFR_c0(Ind2Eli,:,:) = [];
+TFR_c1(Ind2Eli,:,:) = [];
+
+% ----------------------------------------
+% Construct data to keep
+Data.Y              = [Y_c0   ; Y_c1  ];
+Data.Xtfr           = [TFR_c0 ; TFR_c1];
+
+% ----------------------------------------
+% Clear garbage
+clear ans Ind2Eli Ind_c1 Ind_c0 TFR_c0 TFR_c1 Y_c0 Y_c1
+
+
+%% PLOT FOR DEBUGGING
+
+% if (0)
+%     
+%     % ----------------------------------------
+%     for i=1:1:size(Data.Y,1)/2
+%         figure(1)
+%         
+%         subplot(2,1,1)
+%         imagesc(Data.t,Data.f,squeeze(Data.Xtfr(i,:,:)))
+%         xlabel('Time (s)'), ylabel('Frequency (Hz)'), title('n(t)')
+%         colormap jet, view(0,90), box on, grid on, set(gca,'YDir','normal')
+%         
+%         subplot(2,1,2)
+%         imagesc(Data.t,Data.f,squeeze(Data.Xtfr(i+size(Data.Y,1)/2,:,:)))
+%         xlabel('Time (s)'), ylabel('Frequency (Hz)'), title(['n(t)+h(t)   |   SNR=' num2str(Data.Y(i+size(Data.Y,1)/2,5))])
+%         colormap jet, view(0,90), box on, grid on, set(gca,'YDir','normal')
+%         
+%         pause(0.2)
+%         
+%     end % for i=1:1:size(Data.Y,1)/2
+%     clear ans i
+%     
+%     % ----------------------------------------
+%     % Return
+%     return
+% end % if (1)

Compute_ClassificationApply.m

@@ -0,0 +1,13 @@
+function [YEsti, YProb] = Compute_ClassificationApply(XTest,Model)
+
+
+
+%% APPLY CLASSIFIER: APPLY MODEL TO NEW DATA
+
+% Apply classifier. Get predicted class and the probability for each class
+[YEsti,YProb] = classify(Model.net,XTest);
+
+% Convert predictec class from categorical to double
+if iscategorical(YEsti)
+    YEsti = grp2idx(YEsti);
+end

Compute_ClassificationApplyNB.m

@@ -0,0 +1,26 @@
+function [YEstiNB, YProbNB, CostNB] = Compute_ClassificationApplyNB(XTest,Model)
+
+
+
+%% APPLY CLASSIFIER: APPLY MODEL TO NEW DATA
+
+% Size of input data
+Sx = size(XTest,1);
+Sy = size(XTest,2);
+Sz = size(XTest,3);
+Sn = size(XTest,4);
+
+% i) flat images in XTrain, ii) remove dimension of length 1, iii) transpose
+XTest_flat = transpose(squeeze(reshape(XTest,[Sx*Sy,Sz,Sn])));
+
+% Apply classifier. Get predicted class and the probability for each class
+[YEstiNB,YProbNB,CostNB] = predict(Model.NBModel,XTest_flat);
+
+
+% Convert predictec class from categorical to double
+if iscategorical(YEstiNB)
+    YEstiNB = grp2idx(YEstiNB);
+end
+
+% Convert cell array to double array
+YEstiNB = str2num(cell2mat(YEstiNB));

Compute_ClassificationApplySVM.m

@@ -0,0 +1,26 @@
+function [YEstiSVM, YScoreSVM] = Compute_ClassificationApplySVM(XTest,Model)
+
+
+
+%% APPLY CLASSIFIER: APPLY MODEL TO NEW DATA
+
+% Size of input data
+Sx = size(XTest,1);
+Sy = size(XTest,2);
+Sz = size(XTest,3);
+Sn = size(XTest,4);
+
+% i) flat images in XTrain, ii) remove dimension of length 1, iii) transpose
+XTest_flat = transpose(squeeze(reshape(XTest,[Sx*Sy,Sz,Sn])));
+
+% Apply classifier. Get predicted class and the probability for each class
+[YEstiSVM,YScoreSVM] = predict(Model.SVMModel,XTest_flat);
+
+
+% Convert predictec class from categorical to double
+if iscategorical(YEstiSVM)
+    YEstiSVM = grp2idx(YEstiSVM);
+end
+
+% Convert cell array to double array
+YEstiSVM = str2num(cell2mat(YEstiSVM));