go back to previous if condition from an elseif in for loop

I have a big long state-dependent model that is essentially a nevering set of if elseif conditions within a for loop.

Essentially what I want to do is:

opt = [1 2 3 1];

for t = 1:4
  if opt(t) == 1
        disp('ABC')
  elseif opt(t) == 2
        disp('DEF')
  elseif opt(t) == 3
    if rand() < 0.5
        opt(t) == 1
    else
        opt(t) == 2
    end
  end
end

ABC

DEF

ans = logical
   0

ABC

where, if this opt variable is 3, then there's a 50/50 it "turns" into 1 or 2 in the same time step and runs through those outcomes; this way I don't have to rewrite ABC and DEF (because those are really really long)

Answer 1

[Walter Roberson]

You have received some good answers already, but I wanted to show you how to do what you actually asked for.

opt = [1 2 3 1];

for t = 1:4
    while true
        if opt(t) == 1
            disp('ABC')
            break
        elseif opt(t) == 2
            disp('DEF')
            break
        elseif opt(t) == 3
            if rand() < 0.5
                opt(t) = 1;
            else
                opt(t) = 2;
            end
        end
    end
end

ABC

DEF
DEF

ABC

That is, to repeat previous tests, wrap a while around the tests, and break from the while when you no longer want to loop.

Answer 2

[dpb]

opt = [1 2 3 1];

for t = 1:4
  if opt(t) == 3
    opt(t) = (rand()>=0.5)+1;   % set state from 3
  end
  if opt(t) == 1
        disp('ABC')
  elseif opt(t) == 2
        disp('DEF')
  end

If the decision is truly random within the above contstruct with the loop as given, then it would be just as valid to do

opt = [1 2 3 1];

ix3=(opt==3);
opt(ix3)=(rand(1,numel(ix3))>=0.5)+1;   % set state(3) is  50:50
for t = 1:4
  if opt(t) == 1
        disp('ABC')
  elseif opt(t) == 2
        disp('DEF')
  end

Somehow this must not be quite the actual construct?

Answer 3

[Steven Lord]

I would probably turn your disp calls (which I'm assuming are placeholders for what you're actually doing in those cases) into functions that can be called both in the case where opt(t) is 1 and in half the cases where opt(t) is 3. In this case, since those placeholders are short, I can just use anonymous functions. But these could just as easily be "named" functions (implemented as "function outputs = name(inputs), blah blah, end").

option1 = @(t) fprintf("Hello world! Case %d\n", t);
option2 = @(t) fprintf("Goodbye world! Case %d\n", t);
for trials = 1:10
    opt = randi(3);
    switch opt
        case 1
            option1(1)
        case 2
            option2(2)
        otherwise
            if randi(2) == 1
                option1(3)
            else
                option2(3)
            end
    end
end

Hello world! Case 1

Goodbye world! Case 2

Hello world! Case 3

Goodbye world! Case 2
Goodbye world! Case 2

Goodbye world! Case 3
Goodbye world! Case 3

Hello world! Case 1
Hello world! Case 1
Hello world! Case 1

In the "Hello world! Case 3" cases, opt was 3 and then the "coin flip" turned up heads (randi(2) == 1). In the "Goodbye world! Case 3" cases opt was 3 and the coin flip was tails.

[Kitt]

I don't know how well I can turn all of 'ABC' into a function because it's a bunch more nested if statements with coin tosses of over 900 lines of code. Can that be easily turned into a function?

[dpb]

"how well I can turn all of 'ABC' into a function because it's a bunch more nested if statements with coin tosses of over 900 lines of code. Can that be easily turned into a function?"

If you're trying do this alll in a script, it'll never work as intended until and unless you do factor it into separate pieces.

Yes, anything can be turned into a function including your top level script that should just return the desired results to the workplace, not keep everything in the world public.

Look at https://www.mathworks.com/help/matlab/matlab_prog/share-data-between-workspaces.html that discusses scope and passing variables. Using nested functions can also allow you to have acess to global variables inside the functions so you can control individual loop indices, etc., but still operate on one piece of data although note that MATLAB will not create extra copies of arguments but use memory access to the location of the passed-in array so there's really no need to not pass what is needed.

We can illustrate a trivial code that shows the ideas-- if the following is in an m-file, say testit.m,, it will error -- in fact, we can try to run it here to see...

x=rand(5);
func1
Unrecognized function or variable 'x'.
Error in testit>func1 (line 8)
  x(1:3,:)
Error in testit (line 2)

func2
return

function func1
  disp('Func1')
  x(1:3,:)
  x(1,1)=pi;
end

function func2
  disp('Func2')
  x(1:3,:)
end

because the two local functions each have their own private workspace and don't know anything about the base workspace where the variable x lives.

OTOH, if we took the same code but enclosed it all in a function like

testit

Func1

ans = 3×5

0.2388 0.1377 0.0880 0.5657 0.5398 0.4936 0.3527 0.6460 0.3572 0.2075 0.2880 0.7001 0.1960 0.5999 0.2250

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

Func2

ans = 3×5

3.1416 0.1377 0.0880 0.5657 0.5398 0.4936 0.3527 0.6460 0.3572 0.2075 0.2880 0.7001 0.1960 0.5999 0.2250

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

function testit
    x=rand(5);
    func1
    func2
    return
    
    function func1
      disp('Func1')
      x(1:3,:)
      x(1,1)=pi;
    end
    
    function func2
      disp('Func2')
      x(1:3,:)
    end
end

we in fact see that the first functioin did change the first element in the x array to the value of pi; these now are internal functions and by being fully contained inside |function testit|, everything visible in it is visible inside each.

OTOH, if it were like

test2

Func1

ans = 3×5

0.3766 0.8001 0.0680 0.2755 0.6627 0.5545 0.3377 0.0259 0.9714 0.0668 0.2262 0.6714 0.7218 0.5508 0.6005

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

Func2

Unrecognized function or variable 'x'.

Error in solution>func2 (line 51)
x(1:3,:)
^^^^^^^^
Error in solution>test2 (line 39)
func2
^^^^^

function test2
    x=rand(5);
    func1
    func2
    return
    
    function func1
      disp('Func1')
      x(1:3,:)
      x(1,1)=pi;
    end
end

function func2
  disp('Func2')
  x(1:3,:)
end

then func1 is internal and func2 is local and can't see x again. It would need to be called as

function test2
    x=rand(5);
    func1
    func2(x)
    return

Sight unseen, I'd probably opt to start with wrapping your entire code in a function and then make the lines of your standin code internal functions initially. Then I'd almost immediately move each of those outside the main function and pass into them what they need from the main function at the top.