FitWeibTAFCT

1
Hi,
I tried to fit some of my data with FitWeibTAFCT and I keep getting
this:

Warning: Divide by zero.
> In Macintosh HD:MATLAB 5:Toolbox:optim:fminu.m at line 134
In Macintosh HD:MATLAB
5:Toolbox:PsychToolbox:Psychometric:FitWeibTAFC.m at line 59
In Macintosh HD:Desktop Folder:FitWeibullTAFCTest.m at line 17


This only happens for some input numbers. Below is an example of
input levels that don't work [4 5 6 7 8] and input levels that work
[1 2 3 4 5]. Does anybody know why is this happening?

Duje Tadin


% FitWeibTAFCTest

% Set up some input levels
inputs = [4 5 6 7 8]'; % THIS DOESN'T WORK
% inputs = [1 2 3 4 5]'; % THESE NUMBERS WORK JUST FINE

% Cook some data
nCorrect = [52 60 75 90 98]';
nError = 100-nCorrect;

% Do the fit
[alpha,beta,thresh] = FitWeibTAFC(inputs,nCorrect,nError);

% Compute predicted psychometric function
pInputs = 1:0.1:9;
predict = ComputeWeibTAFC(pInputs,alpha,beta);

% Make a plot
hold off
plot(inputs,nCorrect./(nCorrect+nError),'+');
hold on
plot(pInputs,predict,'
2
Dear Duje,

The problem was that FitWeibTAFC (also FitWeibAlphaTAFC, FitWeibAlphaYN)
were making a pretty dumb guess about the initial value of parameter alpha
when such was not supplied initially. The routines were always starting
their search at a value of alpha = 1, and in the second case of input
levels you used, this value was so far off that the numerical search never
converged.

I modified all three functions to initialize alpha to the mean value of the
input levels. This seems like a smarter guess and works for both test
cases you provided. You can also supply a guess based on your knowledge
of the particular data you are trying to fit, but I'm guessing this won't
be necessary with the modified versions.

Thanks for identifying this bug. I've attached the fixed routines, they
will propagate in the next version of the toolbox.

Best,

David
3
I noticed that the CopyWindow routine seems to wait for a screen refresh
before returning when it is copying to the onscreen window (but not to
offscreen windows). Is there some way to disable this? I want to draw
multiple small stimuli to the screen as quickly as possible, and one way
to do this is to copy from an offscreen window to an onscreen window
multiple times rather than redrawing the individual components of each
stimulus multiple times.

I am using:

SCREEN('CopyWindow', w2, w1, rect2, rect1, 'srcCopyQuickly');

Where w2 is an offscreen window and w1 is an onscreen window. This takes
about 11 ms (1 screen refresh), but coping the same image to an offscreen
window takes 3 ms or less.

Is CopyWindow always faster than the SCREEN drawing commands (FillRect,
etc.)?

keith