Optimizing a loop with Kronecker product - matlab

I have three matrices, A_x [nx, np], A_y [ny, np], and A_z [nz, np], and I combine them into an [nx, ny, np] matrix by computing the Kronecker product twice in a loop in Matlab as shown below. Is there a way to do it more efficiently? Ideally, using the GPU and without the loop but any tips would be welcome.
nx = 141;
ny = 141;
nz = 241;
np = 137919;
A_x = rand(nx, np);
A_y = rand(ny, np);
A_z = rand(nz, np);
res = zeros(nx, ny, nz);
for i = 1:np
res = res + reshape(kron(kron(A_x(:,i), A_y(:,i).'), A_z(:,i)), [nx, ny, nz]);
end

Related

FastICA Implementation.. Matlab

I have been working on a FastICA algorithm implementation using MatLab. Currently the code does not separate the signals as good as id like. I was wondering if anyone here could give me some advice on what I could do to fix this problem?
disp('*****Importing Signals*****');
s = [1,30000];
[m1,Fs1] = audioread('OSR_us_000_0034_8k.wav', s);
[f1,Fs2] = audioread('OSR_us_000_0017_8k.wav', s);
ss = size(f1,1);
n = 2;
disp('*****Mixing Signals*****');
A = randn(n,n); %developing mixing matrix
x = A*[m1';f1']; %A*x
m_x = sum(x, n)/ss; %mean of x
xx = x - repmat(m_x, 1, ss); %centering the matrix
c = cov(x');
sq = inv(sqrtm(c)); %whitening the data
x = c*xx;
D = diff(tanh(x)); %setting up newtons method
SD = diff(D);
disp('*****Generating Weighted Matrix*****');
w = randn(n,1); %Random weight vector
w = w/norm(w,2); %unit vector
w0 = randn(n,1);
w0 = w0/norm(w0,2); %unit vector
disp('*****Unmixing Signals*****');
while abs(abs(w0'*w)-1) > size(w,1)
w0 = w;
w = x*D(w'*x) - sum(SD'*(w'*x))*w; %perform ICA
w = w/norm(w, 2);
end
disp('*****Output After ICA*****');
sound(w'*x); % Supposed to be one of the original signals
subplot(4,1,1);plot(m1); title('Original Male Voice');
subplot(4,1,2);plot(f1); title('Original Female Voice');
subplot(4,1,4);plot(w'*x); title('Post ICA: Estimated Signal');
%figure;
%plot(z); title('Random Mixed Signal');
%figure;
%plot(100*(w'*x)); title('Post ICA: Estimated Signal');
Your covariance matrix c is 2 by 2, you cannot work with that. You have to mix your signal multiple times with random numbers to get anywhere, because you must have some signal (m1) common to different channels. I was unable to follow through your code for fast-ICA but here is a PCA example:
url = {'https://www.voiptroubleshooter.com/open_speech/american/OSR_us_000_0034_8k.wav';...
'https://www.voiptroubleshooter.com/open_speech/american/OSR_us_000_0017_8k.wav'};
%fs = 8000;
m1 = webread(url{1});
m1 = m1(1:30000);
f1 = webread(url{2});
f1 = f1(1:30000);
ss = size(f1,1);
n = 2;
disp('*****Mixing Signals*****');
A = randn(50,n); %developing mixing matrix
x = A*[m1';f1']; %A*x
[www,comp] = pca(x');
sound(comp(:,1)',8000)

Vectorize a regression map calculation

I compute the regression map of a time series A(t) on a field B(x,y,t) in the following way:
A=1:10; %time
B=rand(100,100,10); %x,y,time
rc=nan(size(B,1),size(B,2));
for ii=size(B,1)
for jj=1:size(B,2)
tmp = cov(A,squeeze(B(ii,jj,:))); %covariance matrix
rc(ii,jj) = tmp(1,2); %covariance A and B
end
end
rc = rc/var(A); %regression coefficient
Is there a way to vectorize/speed up code? Or maybe some built-in function that I did not know of to achieve the same result?
In order to vectorize this algorithm, you would have to "get your hands dirty" and compute the covariance yourself. If you take a look inside cov you'll see that it has many lines of input checking and very few lines of actual computation, to summarize the critical steps:
y = varargin{1};
x = x(:);
y = y(:);
x = [x y];
[m,~] = size(x);
denom = m - 1;
xc = x - sum(x,1)./m; % Remove mean
c = (xc' * xc) ./ denom;
To simplify the above somewhat:
x = [x(:) y(:)];
m = size(x,1);
xc = x - sum(x,1)./m;
c = (xc' * xc) ./ (m - 1);
Now this is something that is fairly straightforward to vectorize...
function q51466884
A = 1:10; %time
B = rand(200,200,10); %x,y,time
%% Test Equivalence:
assert( norm(sol1-sol2) < 1E-10);
%% Benchmark:
disp([timeit(#sol1), timeit(#sol2)]);
%%
function rc = sol1()
rc=nan(size(B,1),size(B,2));
for ii=1:size(B,1)
for jj=1:size(B,2)
tmp = cov(A,squeeze(B(ii,jj,:))); %covariance matrix
rc(ii,jj) = tmp(1,2); %covariance A and B
end
end
rc = rc/var(A); %regression coefficient
end
function rC = sol2()
m = numel(A);
rB = reshape(B,[],10).'; % reshape
% Center:
cA = A(:) - sum(A)./m;
cB = rB - sum(rB,1)./m;
% Multiply:
rC = reshape( (cA.' * cB) ./ (m-1), size(B(:,:,1)) ) ./ var(A);
end
end
I get these timings: [0.5381 0.0025] which means we saved two orders of magnitude in the runtime :)
Note that a big part of optimizing the algorithm is assuming you don't have any "strangeness" in your data, like NaN values etc. Take a look inside cov.m to see all the checks that we skipped.

Interpolation using polyfit (Matlab)

My script is supposed to run Runge-Kutta and then interpolate around the tops using polyfit to calculate the max values of the tops. I seem to get the x-values of the max points correct but the y-values are off for some reason. Have sat with it for 3 days now. The problem should be In the last for-loop when I calculate py?
Function:
function funk = FU(t,u)
L0 = 1;
C = 1*10^-6;
funk = [u(2); 2.*u(1).*u(2).^2./(1+u(1).^2) - u(1).*(1+u(1).^2)./(L0.*C)];
Program:
%Runge kutta
clear all
close all
clc
clf
%Given values
U0 = [240 1200 2400];
L0 = 1;
C = 1*10^-6;
T = 0.003;
h = 0.000001;
W = [];
% Runge-Kutta 4
for i = 1:3
u0 = [0;U0(i)];
u = u0;
U = u;
tt = 0:h:T;
for t=tt(1:end-1)
k1 = FU(t,u);
k2 = FU(t+0.5*h,u+0.5*h*k1);
k3 = FU((t+0.5*h),(u+0.5*h*k2));
k4 = FU((t+h),(u+k3*h));
u = u + (1/6)*(k1+2*k2+2*k3+k4)*h;
U = [U u];
end
W = [W;U];
end
I1 = W(1,:); I2 = W(3,:); I3 = W(5,:);
dI1 = W(2,:); dI2 = W(4,:); dI3 = W(6,:);
I = [I1; I2; I3];
dI = [dI1; dI2; dI3];
%Plot of the currents
figure (1)
plot(tt,I1,'r',tt,I2,'b',tt,I3,'g')
hold on
legend('U0 = 240','U0 = 1200','U0 = 2400')
BB = [];
d = 2;
px = [];
py = [];
format short
for l = 1:3
[H,Index(l)]=max(I(l,:));
Area=[(Index(l)-2:Index(l)+2)*h];
p = polyfit(Area,I(Index(l)-2:Index(l)+2),4);
rotp(1,:) = roots([4*p(1),3*p(2),2*p(3),p(4)]);
B = rotp(1,2);
BB = [BB B];
Imax(l,:)=p(1).*B.^4+p(2).*B.^3+p(3).*B.^2+p(4).*B+p(5);
Tsv(i)=4*rotp(1,l);
%px1 = linspace(h*(Index(l)-d-1),h*(Index(l)+d-2));
px1 = BB;
py1 = polyval(p,px1(1,l));
px = [px px1];
py = [py py1];
end
% Plots the max points
figure(1)
plot(px1(1),py(1),'b*-',px1(2),py(2),'b*-',px1(3),py(3),'b*-')
hold on
disp(Imax)
Your polyfit line should read:
p = polyfit(Area,I(l, Index(l)-2:Index(l)+2),4);
More interestingly, take note of the warnings you get about poor conditioning of that polynomial (I presume you're seeing these). Why? Partly because of numerical precision (your numbers are very small, scaled around 10^-6) and partly because you're asking for a 4th-order fit to five points (which is singular). To do this "better", use more input points (more than 5), or a lower-order polynomial fit (quadratic is usually plenty), and (probably) rescale before you use the polyfit tool.
Having said that, in practice this problem is often solved using three points and a quadratic fit, because it's computationally cheap and gives very nearly the same answers as more complex approaches, but you didn't get that from me (with noiseless data like this, it doesn't much matter anyway).

MATLAB: 2-D plot with z-axis given in color

My friends and I have been struggling to generate a 2-D plot in MATLAB with
$\eta_1$ and $\eta_2$ both varying in $0:0.01:1$ and the z-axis given by color.
We have a system of 8 differential equations, with HIVinf representing the total new HIV infections in a population over 1 year (HIVinf is obtained by integrating a function of $\eta_1, \eta_2$).
We are looping through $\eta_1$ and $\eta_2$ (two 'for' loops) with the ode45 solver within the 'for' loops.
Based on our prior numerical results, we should be getting much color variation in the 2D-plot. There should be patterns of darkness (high concentration of HIVinfections) along the edges of the plot, and lightness along the diagonals (low concentrations).
However, the following snippet does not produce what we want (I have attached the figure below).
[X,Y] = meshgrid(eta_11,eta_22);
figure;
pcolor(X,Y,AA);
shading interp;
I have attached the code below, as concisely as possible. The function ydot works fine (it is required to run ode45).
We would greatly appreciate if you could help us fix the snippet.
function All()
global Lambda mu mu_A mu_T beta tau eta_1 eta_2 lambda_T rho_1 rho_2 gamma
alpha = 20;
TIME = 365;
eta_11 = zeros(1,alpha);
eta_22 = zeros(1,alpha);
AA = zeros(1,alpha);
BB = zeros(1,alpha);
CC = zeros(1,alpha);
for n = 1:1:alpha
for m = 1:1:alpha
Lambda = 531062;
mu = 1/25550;
mu_A = 1/1460;
mu_T = 1/1825;
beta = 187/365000;
tau = 4/365;
lambda_T = 1/10;
rho_1 = 1/180;
rho_2 = 1/90;
gamma = 1/1000;
eta_1 = (n-1)./(alpha-1);
eta_11(m) = (m-1)./(alpha-1);
eta_2 = (m-1)./(alpha-1);
eta_22(m) = (m-1)./(alpha-1);
y0 = [191564208, 131533276, 2405629, 1805024, 1000000, 1000000, 500000, 500000];
[t,y] = ode45('SimplifiedEqns',[0:1:TIME],y0);
N = y(:,1)+y(:,2)+y(:,3)+y(:,4)+y(:,5)+y(;,6)+y(:,7)+y(:,8);
HIVinf1=[0:1:TIME];
HIVinf2=[beta.*(S+T).*(C1+C2)./N];
HIVinf=trapz(HIVinf1,HIVinf2);
AA(n,m) = HIVinf;
end
end
[X,Y] = meshgrid(eta_11,eta_22);
figure;
pcolor(X,Y,AA);
shading interp;
function ydot = SimplifiedEqns(t,y)
global Lambda mu mu_A mu_T beta tau eta_1 eta_2 lambda_T rho_1 rho_2 gamma
S = y(1);
T = y(2);
H = y(3);
C = y(4);
C1 = y(5);
C2 = y(6);
CM1 = y(7);
CM2 = y(8);
N = S + T + H + C + C1 + C2 + CM1 + CM2;
ydot = zeros(8,1);
ydot(1)=Lambda-mu.*S-beta.*(H+C+C1+C2).*(S./N)-tau.*(T+C).*(S./N);
ydot(2)=tau.*(T+C).*(S./N)-beta.*(H+C+C1+C2).*(T./N)-(mu+mu_T).*T;
ydot(3)=beta.*(H+C+C1+C2).*(S./N)-tau.*(T+C).*(H./N)-(mu+mu_A).*H;
ydot(4)=beta.*(H+C+C1+C2).*(T./N)+tau.*(T+C).*(H./N)- (mu+mu_A+mu_T+lambda_T).*C;
ydot(5)=lambda_T.*C-(mu+mu_A+rho_1+eta_1).*C1;
ydot(6)=rho_1.*C1-(mu+mu_A+rho_2+eta_2).*C2;
ydot(7)=eta_1.*C1-(mu+rho_1+gamma).*CM1;
ydot(8)=eta_2.*C2-(mu+rho_2+gamma.*(rho_1)./(rho_1+rho_2)).*CM2+(rho_1).*CM1;
end
end
Ok, I don't really know much about how the plot should look like, but your eta_11 and eta_22 are variables which are indexed only on the inner loop. That means that when n=1, m=1,2,3,...,alpha your eta_11/eta_22 will be a vector whose elements 1,2,3,...,alpha will be overwritten for every n. Since your meshgrid is outside of the loop, that could be a problem. Usually if you are plotting functions of two variables and you have said variables in 2 nested loops you just ignore the meshgrid. Like this
Option 1:
x=[0:0.01:1];
[x1,x2]=meshgrid(x,x);
y=x1+cos(x2);
contour(x,x,y,30);
Option 2
x=[0:0.01:1];
for i=1:101 %length(x)
for j=1:101
y(i,j)=x1(i)+cos(x2(j)); % It is important to index y with both
% loop variables
end
end
contour(x,x,y,30)

spiral meshgrid in matlab

I'm trying to produce some computer generated holograms by using MATLAB. I used equally spaced mesh grid to initialize the spatial grid, and I got the following image
This pattern is sort of what I need except the center region. The fringe should be sharp but blurred. I think it might be the problem of the mesh grid. I tried generate a grid in polar coordinates and the map it into Cartesian coordinates by using MATLAB's pol2cart function. Unfortunately, it doesn't work as well. One may suggest that using fine grids. It doesn't work too. I think if I can generate a spiral mesh grid, perhaps the problem is solvable. In addition, the number of the spiral arms could, in general, be arbitrary, could anyone give me a hint on this?
I've attached the code (My final projects are not exactly the same, but it has a similar problem).
clc; clear all; close all;
%% initialization
tic
lambda = 1.55e-6;
k0 = 2*pi/lambda;
c0 = 3e8;
eta0 = 377;
scale = 0.25e-6;
NELEMENTS = 1600;
GoldenRatio = (1+sqrt(5))/2;
g = 2*pi*(1-1/GoldenRatio);
pntsrc = zeros(NELEMENTS, 3);
phisrc = zeros(NELEMENTS, 1);
for idxe = 1:NELEMENTS
pntsrc(idxe, :) = scale*sqrt(idxe)*[cos(idxe*g), sin(idxe*g), 0];
phisrc(idxe) = angle(-sin(idxe*g)+1i*cos(idxe*g));
end
phisrc = 3*phisrc/2; % 3 arms (topological charge ell=3)
%% post processing
sigma = 1;
polfilter = [0, 0, 1i*sigma; 0, 0, -1; -1i*sigma, 1, 0]; % cp filter
xboundl = -100e-6; xboundu = 100e-6;
yboundl = -100e-6; yboundu = 100e-6;
xf = linspace(xboundl, xboundu, 100);
yf = linspace(yboundl, yboundu, 100);
zf = -400e-6;
[pntobsx, pntobsy] = meshgrid(xf, yf);
% how to generate a right mesh grid such that we can generate a decent result?
pntobs = [pntobsx(:), pntobsy(:), zf*ones(size(pntobsx(:)))];
% arbitrary mesh may result in "wrong" results
NPNTOBS = size(pntobs, 1);
nxp = length(xf);
nyp = length(yf);
%% observation
Eobs = zeros(NPNTOBS, 3);
matlabpool open local 12
parfor nobs = 1:NPNTOBS
rp = pntobs(nobs, :);
Erad = [0; 0; 0];
for idx = 1:NELEMENTS
rs = pntsrc(idx, :);
p = exp(sigma*1i*2*phisrc(idx))*[1 -sigma*1i 0]/2; % simplified here
u = rp - rs;
r = sqrt(u(1)^2+u(2)^2+u(3)^2); %norm(u);
u = u/r; % unit vector
ut = [u(2)*p(3)-u(3)*p(2),...
u(3)*p(1)-u(1)*p(3), ...
u(1)*p(2)-u(2)*p(1)]; % cross product: u cross p
Erad = Erad + ... % u cross p cross u, do not use the built-in func
c0*k0^2/4/pi*exp(1i*k0*r)/r*eta0*...
[ut(2)*u(3)-ut(3)*u(2);...
ut(3)*u(1)-ut(1)*u(3); ...
ut(1)*u(2)-ut(2)*u(1)];
end
Eobs(nobs, :) = Erad; % filter neglected here
end
matlabpool close
Eobs = Eobs/max(max(sum(abs(Eobs), 2))); % normailized
%% source, gaussian beam
E0 = 1;
w0 = 80e-6;
theta = 0; % may be titled
RotateX = [1, 0, 0; ...
0, cosd(theta), -sind(theta); ...
0, sind(theta), cosd(theta)];
Esrc = zeros(NPNTOBS, 3);
for nobs = 1:NPNTOBS
rp = RotateX*[pntobs(nobs, 1:2).'; 0];
z = rp(3);
r = sqrt(sum(abs(rp(1:2)).^2));
zR = pi*w0^2/lambda;
wz = w0*sqrt(1+z^2/zR^2);
Rz = z^2+zR^2;
zetaz = atan(z/zR);
gaussian = E0*w0/wz*exp(-r^2/wz^2-1i*k0*z-1i*k0*0*r^2/Rz/2+1i*zetaz);% ...
Esrc(nobs, :) = (polfilter*gaussian*[1; -1i; 0]).'/sqrt(2)/2;
end
Esrc = [Esrc(:, 2), Esrc(:, 3), Esrc(:, 1)];
Esrc = Esrc/max(max(sum(abs(Esrc), 2))); % normailized
toc
%% visualization
fringe = Eobs + Esrc; % I'll have a different formula in my code
normEsrc = reshape(sum(abs(Esrc).^2, 2), [nyp nxp]);
normEobs = reshape(sum(abs(Eobs).^2, 2), [nyp nxp]);
normFringe = reshape(sum(abs(fringe).^2, 2), [nyp nxp]);
close all;
xf0 = linspace(xboundl, xboundu, 500);
yf0 = linspace(yboundl, yboundu, 500);
[xfi, yfi] = meshgrid(xf0, yf0);
data = interp2(xf, yf, normFringe, xfi, yfi);
figure; surf(xfi, yfi, data,'edgecolor','none');
% tri = delaunay(xfi, yfi); trisurf(tri, xfi, yfi, data, 'edgecolor','none');
xlim([xboundl, xboundu])
ylim([yboundl, yboundu])
% colorbar
view(0,90)
colormap(hot)
axis equal
axis off
title('fringe thereo. ', ...
'fontsize', 18)
I didn't read your code because it is too long to do such a simple thing. I wrote mine and here is the result:
the code is
%spiral.m
function val = spiral(x,y)
r = sqrt( x*x + y*y);
a = atan2(y,x)*2+r;
x = r*cos(a);
y = r*sin(a);
val = exp(-x*x*y*y);
val = 1/(1+exp(-1000*(val)));
endfunction
%show.m
n=300;
l = 7;
A = zeros(n);
for i=1:n
for j=1:n
A(i,j) = spiral( 2*(i/n-0.5)*l,2*(j/n-0.5)*l);
end
end
imshow(A) %don't know if imshow is in matlab. I used octave.
the key for the sharpnes is line
val = 1/(1+exp(-1000*(val)));
It is logistic function. The number 1000 defines how sharp your image will be. So lower it for more blurry image or higher it for sharper.
I hope this answers your question ;)
Edit: It is real fun to play with. Here is another spiral:
function val = spiral(x,y)
s= 0.5;
r = sqrt( x*x + y*y);
a = atan2(y,x)*2+r*r*r;
x = r*cos(a);
y = r*sin(a);
val = 0;
if (abs(x)<s )
val = s-abs(x);
endif
if(abs(y)<s)
val =max(s-abs(y),val);
endif
%val = 1/(1+exp(-1*(val)));
endfunction
Edit2: Fun, fun, fun! Here the arms do not get thinner.
function val = spiral(x,y)
s= 0.1;
r = sqrt( x*x + y*y);
a = atan2(y,x)*2+r*r; % h
x = r*cos(a);
y = r*sin(a);
val = 0;
s = s*exp(r);
if (abs(x)<s )
val = s-abs(x);
endif
if(abs(y)<s)
val =max(s-abs(y),val);
endif
val = val/s;
val = 1/(1+exp(-10*(val)));
endfunction
Damn your question I really need to study for my exam, arghhh!
Edit3:
I vectorised the code and it runs much faster.
%spiral.m
function val = spiral(x,y)
s= 2;
r = sqrt( x.*x + y.*y);
a = atan2(y,x)*8+exp(r);
x = r.*cos(a);
y = r.*sin(a);
val = 0;
s = s.*exp(-0.1*r);
val = r;
val = (abs(x)<s ).*(s-abs(x));
val = val./s;
% val = 1./(1.+exp(-1*(val)));
endfunction
%show.m
n=1000;
l = 3;
A = zeros(n);
[X,Y] = meshgrid(-l:2*l/n:l);
A = spiral(X,Y);
imshow(A)
Sorry, can't post figures. But this might help. I wrote it for experiments with amplitude spatial modulators...
R=70; % radius of curvature of fresnel lens (in pixel units)
A=0; % oblique incidence by linear grating (1=oblique 0=collinear)
B=1; % expanding by fresnel lens (1=yes 0=no)
L=7; % topological charge
Lambda=30; % linear grating fringe spacing (in pixels)
aspect=1/2; % fraction of fringe period that is white/clear
xsize=1024; % resolution (xres x yres number data pts calculated)
ysize=768; %
% define the X and Y ranges (defined to skip zero)
xvec = linspace(-xsize/2, xsize/2, xsize); % list of x values
yvec = linspace(-ysize/2, ysize/2, ysize); % list of y values
% define the meshes - matrices linear in one dimension
[xmesh, ymesh] = meshgrid(xvec, yvec);
% calculate the individual phase components
vortexPh = atan2(ymesh,xmesh); % the vortex phase
linPh = -2*pi*ymesh; % a phase of linear grating
radialPh = (xmesh.^2+ymesh.^2); % a phase of defocus
% combine the phases with appropriate scales (phases are additive)
% the 'pi' at the end causes inversion of the pattern
Ph = L*vortexPh + A*linPh/Lambda + B*radialPh/R^2;
% transmittance function (the real part of exp(I*Ph))
T = cos(Ph);
% the binary version
binT = T > cos(pi*aspect);
% plot the pattern
% imagesc(binT)
imagesc(T)
colormap(gray)