Struggling with for loop if and else to set up a confusion matrix

Hi there, I have an array of 2 by 100 called Finalprobs, the first row contains results for healthy people and the second row contains results for unhealthy people. I want to set up a test to see if it is possible to differentiate between the healthy and the unhealthy patients by forming a confusion matrix.

I have set a value for K (just as an example here of 1) and if the value for unhealthy patients in the second row of Finalprobs is lower than K I want to add one to a count of TP (true positive), if the value is higher I want to add one to FN(false negative). I want to do this 100 time for all values of unhealthy patients.

For the healthy patients on the first row of Finalprobs I then want to test but here if their value is lower than K one should be added to FP(false positive), if the value is higher here I want to add one to TN(true negative). Again I want to do this 100 times for all values for healthy patients.

At the end I should therefore have 200 counts in total across the 4 parts of the confusion matrix.

Below is the code that I have tried so far by trying to piece together other bits of code from other more simple loops I have used previously, but I am a bit out of my depth here I think and have commented in what I am wanting to do but not sure if the code I have written is close to performing the job.

TP=0
TN=0
FP=0
FN=0        %set variables for the confusion matrix

K = 1       %set K to 1 for testing purposes
    for confuseloopAF = 1:100   %start of for loop for unhealthy patients

              if (Finalprobs(2,:)) < K  %if unhealthy patients value is less
              TP=TP+1;                  %than K add one to count of TP
          else
              if (Finalprobs(2,:)) > K  %if unhealthy patients value is more
              FN=FN+1;                  % than K add one to count of FN
              else
              confuseloopAF;            %run loop again until end
              end;

       for confuseloopCO = 1:100        %start of loop for healthy patients

              if (Finalprobs(1,:)) < K  %if healthy patients value is less 
              FP=FP+1;                  %than K add one to count of FP
          else
              if (Finalprobs(1,:)) < K  % if healthy patients value is more 
              TN=TN+1;                  % than K add one to count of TN
              else
              confuseloopCO;            %run loop again until end
              end;

You will see that I have broken it down into two separate loops as I was really struggling to get my head around putting it all into 1 loop although I know it is possible. It would probably be better in one loop but I was just trying to break it down to as simple as possible in my code to try to get it to work somehow.

Answer 1

[Jan]

The statement "confuseloopAF; %run loop again until end" is confusing only. Simply omit it.

Your loops miss a trailing end command. And the loop counter confuseloopAF is not used anywhere. This looks like the for loop is not clear yet. Please read: doc for.

The code does not need a loop at all:

TP = sum(Finalprobs(2,:) < K)
FN = sum(Finalprobs(2,:) > K)
FP = sum(Finalprobs(1,:) < K)
TN = sum(Finalprobs(1,:) < K)  % Same as FP?!

Instead of sum the nnz can count the number of non-zero elements also.

[Ross Johnston]

Hi Jan, I hadn't seen this answer before posting my other answer, I will give your method a go and see how it pans out for me now as it is more simple than what I have done. (you are correct TN should have been >)

I now need to create a loop that will test for 100 different values for K and save them into a 4 by 100 array for TP,FN,FP and TN so that the data can be analysed, so having the much simpler version that you have created will hopefully be useful for that.

Many thanks!

[Jan]

You do not need a loop over K also. Do you have Matlab 2016b? Then:

K = (1:100).';  % Or any other column vector
TP = sum(Finalprobs(2,:) < K, 2);
FN = sum(Finalprobs(2,:) > K, 2);
FP = sum(Finalprobs(1,:) < K, 2);
TN = sum(Finalprobs(1,:) > K, 2);
Result = [TP, FN, FP, TN];

With older Matlab versions:

TP = sum(bsxfun(@lt, Finalprobs(2,:), K), 2);
FN = sum(bsxfun(@gt, Finalprobs(2,:), K), 2);
FP = sum(bsxfun(@lt, Finalprobs(1,:), K), 2);
TN = sum(bsxfun(@gt, Finalprobs(1,:), K), 2);

What is with the case == K? Perhaps a >= or <= would be better. Or with BSXFUN: @ge or @le.

[Ross Johnston]

Hi Jan, I was able to get your way to work and it is a much simpler way to go for what I was asking.

As discussed above I now want to incorporate a way of changing the K value 100 times to give me 100 columns with values for TP,FN,FP and TN. I want to create an array of 4x100 with each column representing one run of the loop for each value (each column should add up to 200).

I'm not sure if your sum method will work for this when running through multiple values for K, will TP,FN,FP and TN have to be reset to zero at the start of each loop to allow this method to work perhaps?

I have included the calculation for the values for K that I want to test against in the code below along with the for loops I created to get the coding to work for me.

I'm not sure how to incorporate a the K values into the loops to give me the 4x100 array at the end which I am hoping to achieve.

Kmax =max(max(Finalprobs'));
Kmin = 0;
K=linspace(Kmax,Kmin,100) ;

for confuseloopAF = 1:100   

          if (Finalprobs(2,:)) < K  
              TP=TP+1;                  

          else
              FN=FN+1;                  

          end

      end

       for confuseloopCO = 1:100        

           if (Finalprobs(1,:)) < K   
              FP=FP+1;                  

           else
              TN=TN+1;               

           end

       end

If you could give me some pointers in how to achieve this I would really appreciate it, or show how to do it using your more simple method that would be even better!

[Jan]

;-) It seems like you did not get the meaning of my last comment. So let me repeat:

K  = linspace(Kmax, Kmin, 100).';  % Need to be a column vector
TP = sum(bsxfun(@lt, Finalprobs(2,:), K), 2);
FN = sum(bsxfun(@ge, Finalprobs(2,:), K), 2);
FP = sum(bsxfun(@lt, Finalprobs(1,:), K), 2);
TN = sum(bsxfun(@ge, Finalprobs(1,:), K), 2);
Result = [TP, FN, FP, TN];

There is your 100 x 4 matrix. Or with R>=2016b:

K  = linspace(Kmax, Kmin, 100).';  % Need to be a column vector
Result = [sum(Finalprobs(2,:) <  K, 2), ...  % TP
          sum(Finalprobs(2,:) >= K, 2), ...  % FN
          sum(Finalprobs(1,:) <  K, 2), ...  % FP
          sum(Finalprobs(1,:) >= K, 2)];     % TN

But if you want to solve it with a loop, you have to use the loop counter inside the loop. This is still missing. See:

Finalprobs = randi(100, 2, 100);  % Test data
K          = 1:100;
Result     = zeros(numel(K), 4);
for iK = 1:numel(K)
  aK = K(iK);
  TP = 0;
  FN = 0;
  FP = 0;
  TN = 0;
  for iProb = 1:size(Finalprobs, 2)
    if (Finalprobs(2, iProb)) < aK  
      TP = TP + 1;
    else
      FN = FN + 1;                  
    end
    if Finalprobs(1, iProb) < aK   
      FP = FP + 1;
    else
      TN = TN + 1;
    end
  end  % for iProb
  Result(iK, :) = [TP, FN, FP, TN];
end  % for iK

Do you see, that I've used iProb and iK inside the loops? Without this the loop calculates the same data in each iteration.

[Ross Johnston]

Thankyou Jan, this has been a great help!

[Jan]

You are welcome. This was an interesting example to compare a simple loop and the smart vectorization of Matlab. Both have advantages.

Answer 2

[Ross Johnston]

Managed to get it to work using the following code by myself in the end, just putting it up here incase it will help someone else searching for a similar problem...

TP=0;
  TN=0;
  FP=0;
  FN=0;       

K = 1;       
    for confuseloopAF = 1:100   

          if (Finalprobs(2,:)) < K  
              TP=TP+1;                  

          else
              FN=FN+1;                  

          end
               confuseloopAF;
      end

       for confuseloopCO = 1:100        

           if (Finalprobs(1,:)) < K   
              FP=FP+1;                  

           else
              TN=TN+1;               

           end

              confuseloopCO; 
       end