Plotting a fit with excluded data results in error "Subscript indices must either be real positive integers or logicals."

I have fitted some data where I have excluded the first point and now would like to plot it. I'm using the command

    exclude = 1;
    handle = plot(fit,'b',time,data,'+k',exclude,'xr');

From that I get the error

    Subscript indices must either be real positive integers or logicals.
    Error in cfit/plot (line 229)
                     xpoints(outliers),ypoints(outliers),S3,...
    Error in calcFRETDecayDelta_broken (line 558)
    handle = plot(fit,'b',time,data,'+k',exclude,'xr');

Setting

    exclude = 2;

or

    exclude = [1 2];

or typecasting the 1 to a boolean works just fine. I'm reading the exclude vector from a textfile and would like to keep it in the index format there.

So I have now made a workaround by making the index exclude a boolean, which is also accepted by plot:

    ex = zeros(max(exclude),1);
    ex(exclude) = 1;
    exclude = boolean(ex);

The question however is, why does plot not accept a simple 1 for exclusion? This is unnecessary cumbersome.

Answer 1

[dpb]

I can reproduce symptoms here with R2017b; appears to be a bug. I thought perhaps using the named parameter form

handle = plot(fit,'b',time,data,'+k','Exclude',exclude,'xr');

might give the parser enough of a hint but no joy there either.

For a workaround similar to yours, you can keep the index vector as the input and use

exclude=ismember([1:length(time)],exclude);

to build the logical vector on the fly; you could define a function handle to use as

fnPlot=@(f,x,y,ex) plot(f,'b',x,y,'+k',ismember([1:length(x],ex),'xr');

and then call as

h=fnPlot(fit,time,data,exclude);

But, it does appear you're right; there is a bug in the input parser of a surprising-to-find hadn't been caught nature. Report via the official TMW support site.

ADDENDUM

The bug is in

function contains = iContainsOnlyLogicalValues(exclude)
% This function has been written to check how an exclude array has been
% stored for the fit options.  It checks to see whether the array contains
% only ones and zeros, and can serve as a logical array
contains = all(ismember(exclude, [0, 1]));
end

If the input is the point vector 1, then the above function returns true but the exclude variable is double instead of logical and later on that fails in the logical addressing usage.

It seems on surface a fixup would be

contains = all(ismember(exclude, [0, 1]));
contains = contains & numel(exclude)>1;

to make the one-element be converted to logical vector when the test reports a failure that doesn't occur now.

ADDENDUM/ERRATUM

I certainly thought I had tested the case before, but...the above fix fails if exclude is empty--I'd have not thought that

all(ismember(exclude, [0, 1]));

for that case would be TRUE; there certainly are no zeros or ones contained in an empty set, but it does.

Therefore, one needs more robust fix; in the short term without spending serious redesign effort I've added a specific test and just return--

contains = all(ismember(exclude, [0, 1]));   % present TMW code
if isempty(exclude), return, end;            % dpb patch 10Jul18
contains = contains & numel(exclude)>1;      % dpb patch 14Jun18

[dpb]

I did make the above patch here and it definitely fixes the specific problem and very quick testing didn't appear to create any others...I'll submit this thread as reference in bug report.

[dpb]

TMW confirms they also reproduce the bug so hasn't been previously reported/fixed it seems...

[Johann Thurn]

I tried your patch as well and it seems to work. I also only did some quick testing, but I don't think this should cause other problems if I just leave it like that. The function will now return 0, even if exclude = 1 is a boolean, but that should not matter, I think.

Thank you!!

[dpb]

That's TRUE, too...I thought of throwing in an isa() for 'numeric' class, but opted for the shorter version. Probably a cleaner fix is possible, indeed.