Undefined function or variable 'isnan'

I am met with the error "Undefined function or variable 'isnan'" when running this code:

%counter = 1:4
%a is a 220x4 cell
%data is a 220x4 double

for i = size(a, 1)
    pairs = [];
    pairs(:,1) = counter;
    pairs(:,2) = data(i,:);
    pairs = pairs(~isnan(pairs(:,2),:));
    slope(i) = polyfit(pairs(:,1),pairs(:,2),1);
end

I can't seem to figure out exactly why. Is the way I'm using the NOT operator?

[Stephan]

What is the result of:

which isnan -all

[Carver Nabb]

[nanren888]

As Walter says; It is complaining about the two-parameter isnan(x,y) that you've invented :)

Answer 1

[Walter Roberson]

You have

pairs = pairs(~isnan(pairs(:,2),:));

You want

pairs = pairs(~isnan(pairs(:,2)),:);

[nanren888]

Yes, it is complaining about the two-parameter isnan(x,y) that you've invented :)

[per isakson]

How difficult could it be to produce a better error message?

>> isnan( 17, : )
Undefined function or variable 'isnan'. 

[Carver Nabb]

Thank you Walter!

[Carver Nabb]

I do have a question about an error with this same code, but about the line:

slope(i) = polyfit(pairs(:,1),pairs(:,2),1);

Would it be best to create a new question? It is telling me the indices on the left side are not compatible with the right.

[Walter Roberson]

It is an oddity of the parser, which has some messy behaviour for colon.

functionname(argument, :)

will disable recognition of functionname as a function and will look for a variable with that name, probably resulting in the misleading error message.

A = reshape(1:60, 2, []);
A(1:2)
%1:2 gets computed as [1 2] first, then the datatype of A is looked up
%and since it is a variable, subsref is invoked to select the first two elements
A('1:2')
%'1:2' is a vector of characters, equivalent to char([49 58 50]) .
%The numeric equivalent is taken, [49 58 50] and are used as subscripts into A
A(:)
%the : is transformed into subsref() with type '()' and index {':'}
%which is treated by subsref as a reshape request
A(':')
%the ':' with the character scalar ':' is transformed into subsref() with
% type '()' and index {':'} which is treated as a reshape request
A(1, :)
%this is transformed into subsref() with type '()' and index {1, ':'}
%which is treated by subsref as a subset indexing request
A(1, ':')
%this is transformed into subsref() with type '()' and index {1, ':'}
%which is treated by subsref as a subset indexing request
A(1, '1:2')
%this is transformed into subsref() with type '()' and index {1, '1:2'}
%which is treated by subsref as a subset indexing request with
%the '1:2' being treated as equivalent to [49 58 50]

but

functionname(argument, ':')

is not treated as a subsref request.

[Walter Roberson]

polyfit() with third parameter N requests a fitting in which the maximum exponent used for the independent variable is N. Since exponents all the way down to 0 are used, that results in N+1 results. So when you requested polyfit(x,y,1) you are going to get out 1+1 = two coefficients, one for x^1 and one for x^0 (the constant term)

[Carver Nabb]

So, just so I'm clear, I need to change the exponent to 0, as well as place the colons in ' ' ? I apologize if this is a silly question, I am quite new to programming.

[Carver Nabb]

When I run the script, after changing the exponent to 0, it runs without error however all values for slope are 0 as well.

[Walter Roberson]

For the isnan() matter, you just accidentally had the : inside the isnan function call and needed to move it outside that call into a place it was acting as a subscript.

polyfit(pairs(:,1),pairs(:,2),0)

is always exactly equal to mean(pairs(:,2)) even when the x are irregularly spaced. If you want only a single coefficient then there is no point using polyfit().

If you want to fit the data to a straight line then you would need to expect two coefficients as output, like

mc = polyfit(pairs(:,1),pairs(:,2),1);
slope(i) = mc(1);

[Jan]

@Carver Nabb: Use:

pairs = pairs(~isnan(pairs(:,2)),:);

Again the debugger will help you to understand, what happens. dbstop if error, run the code again, and when it stops:

polyfit(pairs(:,1),pairs(:,2),1)
size(polyfit(pairs(:,1),pairs(:,2),1))
size(slope(i))

You cannot assign a vector to a scalar. I guess you want:

P = polyfit(pairs(:,1),pairs(:,2),1);
slope(i) = P(2);

[Carver Nabb]

I am still getting Zeros for all of the slope outputs. Here is my revised code. It runs without prompting any errors. Thank you again for the help, I greatly appreciate it.

for i = size(a, 1)
    pairs = [];
    pairs(:,1) = counter;
    pairs(:,2) = data(i,:);
    pairs = pairs(~isnan(pairs(:,2)),:);
    P = polyfit(pairs(:,1),pairs(:,2),1);
    slope(i) = P(1); %I also tried with P(2)
end

[Walter Roberson]

Question: why is your for loop to size(a,1) but you are not accessing a in the loop? It would look to me to make more sense to run to size(data,1)

[Carver Nabb]

Agreed. I've already made the change; a was a cell that i changed to a matrix (data) using cell2mat. They're the same size so it didn't make much of a difference.

[Walter Roberson]

Are you happening to see warnings such as

Warning: Polynomial is not unique; degree >= number of data points.

?

[Carver Nabb]

No, the only tip I see is "The variable 'slope' appears to change size on every loop iteration. Consider preallocating for speed.

[Walter Roberson]

Can you attach a .mat with counter and data ?

[Carver Nabb]

Here you are. Thanks for the help, Walter.

[Carver Nabb]

It looks like the allpcompst.mat I've sent you already has the NaN values. So a part of the code is unnecessary.

[Walter Roberson]

for i = size(allpcompst,1)

should be

for i = 1:size(allpcompst,1)

You will get messages Warning: Polynomial is not unique; degree >= number of data points. 207 of the 220 slopes are non-zero.

[Carver Nabb]

Thank you, this does produce non-zero values. However, they are not the correct slopes, for example, I know the first slope ought to be 3.6, and the last, -4.6. However the first value returned with this code is 113.

[Carver Nabb]

I figured it out! Walter and Jan, thank you so much for the help. I do appreciate it.

Answer 2

[Jan]

Although I assume that Walter hit the point already, I suggest to use teh debugger to identify such problems. Type in the command window:

dbstop if error

Now run the code again until it stops. Check the currently processed line by evaluating it piecewise:

% pairs = pairs(~isnan(pairs(:,2),:))
pairs(:,2)
~isnan(pairs(:,2),:)  % <- This should fail already
pairs(~isnan(pairs(:,2),:))

As Walter has said already, isnan(x, :) is meaningful only, if it is an array. Otehrwise using the colon as second input is not valid, if the function isnan() is meant. The error message "Undefined function or variable 'isnan'" is not really clear here. But splitting the commands into parts using the debugger helps to understand the problem.

Answer 3

[Kholoud Alzoubi]

Undefined function or variable 'textdata'.

Error in KopackRFNNGOA (line 15)

x1=textdata(~isnan(textdata)); %temperature

[Walter Roberson]

Not enough context.

You are trying to use a variable named textdata that has not been assigned to.

Also, the variable name would tend to suggest that the variable is not expected to hold numeric data, so isnan() is questionable here.