How to set which part of the code to comment using a variable?

Good morning,

I have a code that uses different algorithms on the same data and I have to compare them.

I want to do the following thing

Flag1=1;
Flag2=0;
Flag3=1;

if Flag1
    DataOut1=zeros(1e9,1);
end
if Flag2
    DataOut2=zeros(1e9,1);
end
if Flag3
    DataOut3=zeros(1e9,1);
end

ParamIn=1:1e9;

for n=1:1e9    
    
    DataIn=randn(1);
    if Flag1
        DataOut1(n)=Algorithm1(ParamIn(n),DataIn);
    end
    if Flag2
        DataOut2(n)=Algorithm2(ParamIn(n),DataIn);
    end
    if Flag3
        DataOut3(n)=Algorithm3(ParamIn(n),DataIn);
    end
    
end

Is there a way to avoid using the if statement and to use the flags to comment the different parts of the code instead. That would save the time to check the if condition since the number of iterations in the for cycle can be really high. I cannot split the for cycles since they have to work on the same DataIn.

Thank you for your help.

Francesco

Answer 1

[SALAH ALRABEEI]

See this example

clear
clc
a=2;
b=3
eval(['c',num2str(a),'=',','A',num2str(a),'(',num2str(b),')'])
function y=A1(x)
  y=sin(x);
end
function y=A2(x)
  y=cos(x);
end

In ur case a=1 or 2 or 3 which is the flag,

     DataOut1(n)= eval(['DataOut',num2str(a),'(n)=',
'Algorithm',num2str(a),'(ParamIn(n),DataIn)'])

[Walter Roberson]

Using eval() is not recommended!

http://www.mathworks.com/matlabcentral/answers/304528-tutorial-why-variables-should-not-be-named-dynamically-eval

[SALAH ALRABEEI]

@Walter yes, i forgot about it.

Answer 2

[Walter Roberson]

If practical, vectorize the Algorithm* routines, and combine them into one routine that also takes the flags.

DataIn = randn(1e9,1);
[Dataout1, Dataout2, Dataout3] = Algorithm(DataIn, Flag1, Flag2, Flag3);

You would not need ParamIm as it would be implied by the row number of DataIn.

Answer 3

[John D'Errico]

NO. You do NOT want to comment out code programmatically. That will NOT be faster, even if you can find a way to do it. Your promatically modified code will now become incredibly slow. (And probably buggy too.) And you want to do this a billion times? Sorry, but you need to learn MATLAB better, learning to write efficient code. But this approach you think you want is NOT the solution.

For example, a simple re-arrangement of your code will perform the test on flag only once for each flag.

Flag1=1;
Flag2=0;
Flag3=1;

if Flag1
    DataOut1=zeros(1e9,1);
    for n=1:1e9    
      DataIn=randn(1);
      DataOut1(n)=Algorithm1(ParamIn(n),DataIn);
    end
end
if Flag2
    DataOut2=zeros(1e9,1);
    for n=1:1e9    
      DataIn=randn(1);
      DataOut1(n)=Algorithm2(ParamIn(n),DataIn);
    end
end
if Flag3
    DataOut2=zeros(1e9,1);
    for n=1:1e9    
      DataIn=randn(1);
      DataOut3(n)=Algorithm3(ParamIn(n),DataIn);
    end
end

It is possible that better code yet (that understands what the various algorithms are) may be eble to completely replace the loops, or at least partially replace them.

[Walter Roberson]

The user's current code uses the same randn() value for all active algorithms; the modified version uses different values. You would need to generate the 1e9 randn ahead of time and feed the appropriate entry into the function. (For space reasons you might want to run batches instead of storing all 1e9 random numbers.)

Answer 4

[Image Analyst]

Once you've got that figured out then, to COMPARE the algorithms, you might want to look into an ROC analysis using the MATLAB function perfcurve().

https://en.wikipedia.org/wiki/Receiver_operating_characteristic