I am trying to visualize a small simulation I have to do. For the task, it is not necessary but as it quite often happens: I run into the problem and want the solution before I go on.
The simulation is an ultra-simple neural net from the computational neuroscience department with an 'Oja' algorithm. I have the value plotted as scatterplot and wanted to animate the changing weight over the loops. That for itself works fine (figure 2 or f2 handle) Then I decided to print the difference in the weight vector from one run to another calculated as the norm. I wanted this to be plotted as a line that evolves over time (figure 1 aka f1). But though I always activate figure 2 it switches back to figure 1 and plots it there.
Of course, I searched the internet as well as stackexchange where btw I found lots of fascinating stuff about animations. Just nothing that solved the problem...
Two questions:
Thanks.
Here is the code:
function [t, w, dw]=weight(X)
w=rand(2,1)*5; %Initialization
%constants
n=1;
alpha=1;
dt=0.01;
T=5;
L=length(X);
w_old=[0; 0];
s=size(w_old)
t=0;
limit=0.001;
%handles for the figures Error and weight animation
f1= figure
set(f1,'DoubleBuffer','on', 'Name','Error');
ax1 = axes('Position',[0.1 0.1 0.7 0.7]);
f2=figure
set(f2, 'Name', 'weights');
%normalizing and plot
X=[X(:,1)-mean(X(:,1)),X(:,2)-mean(X(:,2))];
scatter(X(:,1),X(:,2));
%function handle for the error and the weights animation
herror = animatedline('Marker','.');
hLine = line('XData',w(1), 'YData',w(2), 'Color','r', ...
'Marker','o', 'MarkerSize',6, 'LineWidth',2);
hTxt = text(w(1), w(2), sprintf('(%.3f,%.3f)',w(1),w(2)), ...
'Color',[0.2 0.2 0.2], 'FontSize',8, ...
'HorizontalAlignment','left', 'VerticalAlignment','top');
while (t<T)
for i=1:L
w_old= w;
u=X(i,:);
v=u*w;
w=w+dt*n*(v*u'-alpha*v^2*w); %Oja rule
figure(f2);
hold on;
set(hLine, 'XData',w(1), 'YData',w(2))
set(hTxt, 'Position',[w(1) w(2)], ...
'String',sprintf('(%.3f,%.3f,%.2f)',[w(1) w(2) t]))
drawnow %# force refresh
%#pause(DELAY)
hold off;
dw=norm(w_old-w);
figure(f1)
hold on;
addpoints(herror, (i*t/dt),dw)
drawnow
hold off;
if dw<limit, break; end
end
t=t+dt;
if ~ishandle(hLine), break; end
end
end
You created a new axes ax1 overlapping to the plot hiding the animation.
You also put herror = animatedline('Marker','.'); after f2=figure doing nothing in the first figure before the for loop. This is the working code:
function [t, w, dw]=weight(X)
X = randn(20,2);
w=rand(2,1)*5; %Initialization
close all
%constants
n=1;
alpha=1;
dt=0.01;
T=5;
L=length(X);
w_old=[0; 0];
s=size(w_old);
t=0;
limit=0.001;
%handles for the figures Error and weight animation
f1= figure(1);
hold on;
set(f1,'DoubleBuffer','on', 'Name','Error');
%ax1 = axes('Position',[0.1 0.1 0.7 0.7]);
%function handle for the error and the weights animation
herror = animatedline('Marker','.');
xlabel('t'); ylabel('Error');
f2=figure(2);
hold on;
set(f2, 'Name', 'weights');
xlabel('W1'); ylabel('W2')
%normalizing and plot
X=[X(:,1)-mean(X(:,1)),X(:,2)-mean(X(:,2))];
scatter(X(:,1),X(:,2));
hLine = line('XData',w(1), 'YData',w(2), 'Color','r', ...
'Marker','o', 'MarkerSize',6, 'LineWidth',2);
hTxt = text(w(1), w(2), sprintf('(%.3f,%.3f)',w(1),w(2)), ...
'Color',[0.2 0.2 0.2], 'FontSize',8, ...
'HorizontalAlignment','left', 'VerticalAlignment','top');
while (t<T)
for i=1:L
w_old= w;
u=X(i,:);
v=u*w;
w=w+dt*n*(v*u'-alpha*v^2*w); %Oja rule
set(hLine, 'XData',w(1), 'YData',w(2))
set(hTxt, 'Position',[w(1) w(2)], ...
'String',sprintf('(%.3f,%.3f,%.2f)',[w(1) w(2) t]))
drawnow %# force refresh
dw=norm(w_old-w);
figure(f1);
addpoints(herror, (i*t/dt),dw)
drawnow
if dw<limit; break; end
end
t=t+dt;
if ~ishandle(hLine), break; end
end
end
To me, it seems more natural using just one window and 2 subplots instead of switching window back and forth.