cleanup

noblec04 · Oct 15, 2024 · e8d0f62 · e8d0f62
1 parent 946917d
commit e8d0f62
Show file tree

Hide file tree

Showing 7 changed files with 136 additions and 136 deletions.
diff --git a/MatlabGP/+NN/NN.m b/MatlabGP/+NN/NN.m
@@ -117,6 +117,45 @@
 
         end
 
+        function [e,de] = Batchloss(obj,V,x,y,N)
+
+            nV = length(V(:));
+
+            V = AutoDiff(V(:));
+
+            obj = obj.setHPs(V(:));
+
+            M = 0;
+
+            e = 0;
+            de = 0;
+
+            while size(x,1)>0
+                    M=M+1;
+                    itrain = randsample(size(x,1),min(N,size(x,1)));
+
+                    xt = x(itrain,:);
+                    yt = y(itrain,:);
+
+                    x(itrain,:)=[];
+                    y(itrain,:)=[];
+
+                    [yp] = obj.forward(xt);
+
+                    [eout] = obj.lossfunc.forward(yt,yp);
+
+                    e1 = sum(eout,2);
+
+                    e = e + getvalue(e1);
+                    de1 = getderivs(e1);
+                    de = de + reshape(full(de1),[1 nV]);
+            end
+
+            e = e/M;
+            de = de/M;
+
+        end
+
         function [obj,fval] = train(obj,x,y,lb,ub)%,xv,fv
 
             obj.X = x;
@@ -137,6 +176,34 @@
             func = @(V) obj.loss(V,x,y);
 
 
+            opts = optimoptions('fmincon','SpecifyObjectiveGradient',true,'MaxFunctionEvaluations',3000,'MaxIterations',3000,'Display','final');
+            [theta,fval] = fmincon(func,tx0,[],[],[],[],[],[],[],opts);
+
+            %[theta,fval,xv,fv] = VSGD(func,tx0,'lr',0.01,'gamma',0.01,'iters',1000,'tol',1*10^(-7));
+
+            obj = obj.setHPs(theta(:));
+        end
+
+        function [obj,fval] = Batchtrain(obj,x,y,N,lb,ub)%,xv,fv
+
+            obj.X = x;
+            obj.Y = y;
+
+            if nargin<5
+                obj.lb_x = min(x);
+                obj.ub_x = max(x);
+            else
+                obj.lb_x = lb;
+                obj.ub_x = ub;
+            end
+
+            x = (x - obj.lb_x)./(obj.ub_x - obj.lb_x);
+
+            tx0 = (obj.getHPs());
+
+            func = @(V) obj.Batchloss(V,x,y,N);
+
+
             opts = optimoptions('fmincon','SpecifyObjectiveGradient',true,'MaxFunctionEvaluations',3000,'MaxIterations',3000,'Display','final');
             [theta,fval] = fmincon(func,tx0,[],[],[],[],[],[],[],opts);
 

diff --git a/MatlabGP/+kernels/ICD.asv b/MatlabGP/+kernels/ICD.asv
diff --git a/MatlabGP/examples/TestHetNoiseKernel.m b/MatlabGP/examples/TestHetNoiseKernel.m
@@ -0,0 +1,17 @@
+
+a = kernels.EQ(1,0.25);
+
+H = diag(log([1 2 1 0.5 0.1 0.6 0.9 2]));
+
+X = lhsdesign(200,2);
+x = lhsdesign(8,2);
+
+kxX = a.build(x,X);
+
+K = diag(diag(exp(kxX'*H*kxX)));
+
+figure(1)
+clf(1)
+plot3(x(:,1),x(:,2),diag(exp((H))),'x','MarkerSize',10,'LineWidth',3)
+hold on
+plot3(X(:,1),X(:,2),diag(K),'x','MarkerSize',10,'LineWidth',3)
diff --git a/MatlabGP/examples/TestICD.asv b/MatlabGP/examples/TestICD.asv
diff --git a/MatlabGP/examples/TestICD_2D.asv b/MatlabGP/examples/TestICD_2D.asv
diff --git a/MatlabGP/examples/testNN_2D_batch.m b/MatlabGP/examples/testNN_2D_batch.m
@@ -0,0 +1,40 @@
+
+clear
+close all
+clc
+
+nD = 5;
+nT = 1000;
+
+xx = -2 + 4*[lhsdesign(nT,nD);utils.HypercubeVerts(nD)];
+
+[yy] = testFuncs.Rosenbrock(xx,1)/7210;
+
+xmesh = -2 + 4*lhsdesign(5000,nD);
+ymesh = testFuncs.Rosenbrock(xmesh,1)/7210;
+
+layers{1} = NN.FF(nD,nD);
+layers{2} = NN.FF(nD,6);
+layers{3} = NN.FF(6,1);
+
+acts{1} = NN.SNAKE(8);
+acts{2} = NN.SNAKE(8);
+
+lss = NN.MSE();
+
+nnet = NN.NN(layers,acts,lss);
+
+%%
+
+tic
+[nnet2,fval] = nnet.Batchtrain(xx,yy,20);%,xv,fv
+toc
+
+%%
+
+yp2 = nnet2.predict(xmesh);
+
+%%
+
+figure
+plot(yp2,ymesh,'.')
diff --git a/MatlabGP/examples/testNN_class.asv → MatlabGP/examples/testNN_batch.m b/MatlabGP/examples/testNN_class.asv → MatlabGP/examples/testNN_batch.m
@@ -3,39 +3,33 @@
 close all
 clc
 
-xx = [0;lhsdesign(10,1);1];
+xx = [0;lhsdesign(1000,1);1];
 yy = normrnd(forr(xx,0),0*forr(xx,0)+0.4);
 
-yc(:,1) = double(yy>0);
-yc(:,2) = double(yy<=0);
-
-
 xmesh = linspace(0,1,100)';
 ymesh = forr(xmesh,0);
 
-layers{1} = NN2.FF(1,3);
-layers{2} = NN2.FF(3,6);
-layers{3} = NN2.FF(6,2);
+layers{1} = NN.FF(1,3);
+layers{2} = NN.FF(3,6);
+layers{3} = NN.FF(6,3);
 
-acts{1} = NN2.SNAKE(2);
-acts{2} = NN2.SNAKE(1);
+acts{1} = NN.SNAKE(2);
+acts{2} = NN.SNAKE(1);
 
-lss = NN2.CE();
+lss = NN.MAE();
 
-nnet = NN2.NN(layers,acts,lss);
+nnet = NN.NN(layers,acts,lss);
 
 %%
 
 tic
-[nnet2,fval] = nnet.train(xx,yc);%,xv,fv
+[nnet2,fval] = nnet.Batchtrain(xx,yy,10);%,xv,fv
 toc
 
 %%
 
 yp2 = nnet2.predict(xmesh);
 
-yp3 = exp(yp2)./sum(exp(yp2));
-
 
 %%
 % figure
@@ -64,6 +58,9 @@
     else
         y(i,1) = (6*x(i)-2).^2.*sin(12*x(i)-4)+dx;
     end
+
+    y(i,2) = 0.4*(6*x(i)-2).^2.*sin(12*x(i)-4)-x(i)-1;
+    y(i,3) = A*(6*x(i)-2).^2.*sin(12*x(i)-4)+B*(x(i)-0.5)-C;
 end
 
 end