From raudys@ktl.mii.lt Fri Nov 15 14:59:18 2002 Return-Path: X-Sieve: cmu-sieve 2.0 Received: from Julius.ktl.mii.lt (julius.ktl.mii.lt [193.219.50.3]) by mail.microlink.lt (MicroLink Data mail) with ESMTP id EE331113275; Fri, 15 Nov 2002 14:59:18 +0200 (EET) Received: from NEURONAS (raudys.ktl.mii.lt [193.219.51.18]) by Julius.ktl.mii.lt (8.12.0.Beta16/8.12.0.Beta16) with SMTP id gAFCsZoR030698; Fri, 15 Nov 2002 14:54:36 +0200 Message-ID: <011c01c28ca6$38d5ca30$1233dbc1@NEURONAS> From: =?windows-1257?Q?=D0arunas_Raudys?= To: "Rita" Cc: "Mantas Krakys" , "Robert" , , , "Alexander" , , , , , , "Iljuxa" , , , , "Ramunas Lukasevicius" , , , , , "Marijus Kasperavicius" , , , Subject: Fw: pabandykit prognozuoti Londono birza Date: Fri, 15 Nov 2002 14:54:59 +0200 MIME-Version: 1.0 Content-Type: multipart/mixed; boundary="----=_NextPart_000_0119_01C28CB6.F7F6D340" X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 6.00.2800.1106 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2800.1106 Status: RO This is a multi-part message in MIME format. ------=_NextPart_000_0119_01C28CB6.F7F6D340 Content-Type: text/plain; charset="windows-1257" Content-Transfer-Encoding: 8bit > o noras atlikti Jusu paskirta uzduoti yra didelis... pagirtinas noras bandyk, klausiant norodyk kokia diagnostika. ----- Original Message ----- From: "Ðarunas Raudys" To: "Alexander" Cc: "Mantas Krakys" ; "Robert" ; ; ; "Alexander" ; ; Sent: Monday, October 14, 2002 7:30 AM Subject: Fw: pabandykit prognozuoti Londono birza > > ----- Original Message ----- > From: "Ðarunas Raudys" > To: "Marijus Kasperavicius" > Cc: ; ; > ; ; ; > ; ; "Ramunas Lukasevicius" > ; ; > ; ; > > Sent: Friday, October 04, 2002 1:50 PM > Subject: pabandykit prognozuoti Londono birza > > > > Pradziai reik tureti matlaba suisinstaliavus pas > > save!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! > > > > > > "pratyboms" 13-ka adresu (tiek susirase!) siunciu 3.5 metu Londono > birzos > > duomenis, kur reik prognozuoti x1 indeksa 1 dienai i prieki. > > > > Uzsikraunat duomenis "birza" ir 3 programeles ir leidziate: > > > > > [ER,yt,y_prognozeT,ym,y_prognozeM]=LondBirzSINUS(100,400,500,600,[1,4,10,11, 12,20],20,0.1,3,1,0.5); > > > > Paskui analizuokite rezultatus, bandykit kaitalioti po truputi, kad > > suprastumet, kas yra kas. > > > ------=_NextPart_000_0119_01C28CB6.F7F6D340 Content-Type: application/octet-stream; name="linearSLP.M" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="linearSLP.M" % Find robust regression by % nonlinear single layer perceptron % author Sarunas Raudys % A input N*p array - training-set % Y target N*1 array- training-set % At input Nt*p array - test-set % Yt target Nt*1 array- test-set % iter - a number of iterations % step - learning-step % Wstart - 1*(p+1) starting weight vector % alfa - a scaling parameter % W - 1*(p+1) final weight vector % et - generalization error history in "iter" training iterations % prior to training we recommend: % - to substract from A,Y, and At,Yt the sample means of A, Y, % - to use Wstart = zeros(1,p+1); function [W,Wbest,er,et]=linearSLP(A,Y,At,Yt,iter,step,Wstart) [N , p ] = size(A );[Nt, pt] = size(At);step=2*step/N; W=Wstart;Wbest=Wstart;emin=999; AA=[A, ones(N,1)]';AAt=[At, ones(Nt,1)]';Y=Y';Yt=Yt'; for i=1:iter W=W+step*(Y- W*AA)*AA'; dt=W*AAt-Yt;dr=W*AA-Y; et(i)=sqrt( dt*dt'./Nt);if et(i) % A input N*p array - training-set % Y target N*1 array- training-set % At input Nt*p array - test-set % Yt target Nt*1 array- test-set % iter - a number of iterations % step - learning-step % Wstart - 1*(p+1) starting weight vector % alfa - a scaling parameter % W - 1*(p+1) final weight vector % et - generalization error history in "iter" training iterations % prior to training we recommend: =20 % - to substract from A,Y, and At,Yt the sample means of A, Y,=20 % - to use Wstart =3D zeros(1,p+1); % = figure(1);x=3D[-3.2:0.1:3.2];plot(x,(1-cos(x))/2);figure(2);x=3D[-3.2:0.1= :3.2];plot(x,sin(x)); function = [WBEST,ER,ITER]=3Drobslpsinus(A,Y,iter,step,Wstart,percent,alfa) W=3DWstart;[N , p ] =3D size(A );AA=3D[A, = ones(N,1)];Nperc=3D1+fix(N*percent/100);step=3Dstep/N; for j1=3D1:Nperc [Dev,id1]=3D sort(abs(Y'-W*AA'));maxi=3Dalfa*Dev(end-j1+1);er=3D[]; scale=3D 0.5*pi/maxi;Wbest=3DWstart;emin=3D999; =20 for i1=3D1:iter dist=3DY- AA * W'; ind=3Dfind(abs(dist) % A input N*p array - training-set % Y target N*1 array- training-set % At input Nt*p array - test-set % Yt target Nt*1 array- test-set % iter - a number of iterations % step - learning-step % Wstart - 1*(p+1) starting weight vector % alfa - a scaling parameter % W - 1*(p+1) final weight vector % et - generalization error history in "iter" training iterations % prior to training we recommend: =20 % - to substract from A,Y, and At,Yt the sample means of A, Y,=20 % - to use Wstart =3D zeros(1,p+1); % = figure(1);x=3D[-3.2:0.1:3.2];plot(x,(1-cos(x))/2);figure(2);x=3D[-3.2:0.1= :3.2];plot(x,sin(x)); function = [W,Wbest,er,et,ITER]=3Dpercrobsin(A,Y,At,Yt,iter,step,Wstart,maxi) [N , p ] =3D size(A ); [Nt, pt] =3D size(At); W=3DWstart;Wbest=3DWstart;emin=3D999; =20 AA=3D[A, ones(N,1)];AAt=3D[At, ones(Nt,1)]; scale=3D 0.5*pi/maxi; %figure(3);hist(prm,15);disp([sigmB,prm(n2),prm(n1)]) for i1=3D1:iter dist=3DY- AA * W'; ind=3Dfind(abs(dist)