KNN with parameter tuning and LOG process....Not getting confusion matrix for all k

vbsingh

I am using KNN for multiclass problem . I am applying optimized grid search  for parameter tuning and using log process for getting every cucle results. But I am getting only accuracy results and different values of K. How can i get the confussion matrix  for all possible values of K in KNN..Please suggest and guide...

Telcontar120

This isn't possible with the log operator and optimize operator alone, I think you are going to need to use a loop and store the confusion matrix from each model run separately.

varunm1

Hello @vbsingh

If you are just optimizing on K-Value, please look at the below process, where there are a couple of macros and store operators to store performance in the repository folder. This way, you can store the performance of each k value. You can import the process attached to your RM and change the store operator repository entry, but be sure to add the macro name in the current process of the store operator in the name.

varunm1

@vbsingh by the way, uncheck the "enable parallel execution" in optimize parameter operator.

vbsingh

How to get confusion matrix  for all k values through grid search...

Telcontar120

This isn't possible with the log operator and optimize operator alone, I think you are going to need to use a loop and store the confusion matrix from each model run separately.

vbsingh

But I need parameter optimization too...

varunm1

Hello @vbsingh

If you are just optimizing on K-Value, please look at the below process, where there are a couple of macros and store operators to store performance in the repository folder. This way, you can store the performance of each k value. You can import the process attached to your RM and change the store operator repository entry, but be sure to add the macro name in the current process of the store operator in the name.

vbsingh

Great I am trying it......

varunm1

@vbsingh by the way, uncheck the "enable parallel execution" in optimize parameter operator.

vbsingh

plz send the .properties file also

varunm1

Here is the zip folder with all the related files, you just need to unzip in your local repository of rapidminer folder and run it.

vbsingh

Hi varunm1,
Thanks for sharing your operator.. It was great help. But i have one problem with J48 since there is two parameter ...so there is overlapping in result.. how to change the name of the result..and get entire result.

vbsingh

What should be the range for J48 machine learning technique while using optimize operator