Is it possible to create our own machine learning models and use them in training/prediction process

divya_das

Hi,

Is it possible to create our own machine learning models and use them in training/prediction process?

In case we are not able to achieve the desired accuracy with the machine learning algorithms that RapidMiner provides, can we go ahead and use some other algorithm (e.g. from open source). Will we have to bundle the new algorithm in an operator and then use it in the training process? Or can we invoke it using the execute program/execute script operator? Or is there some other way to use a new algorithm within the training process?

In the RapidMiner documentation, I saw an example where ‘Execute Python’ operator is being used to build our own ML model.  

Regards,

Divya

 

rfuentealba

Hello Divya,

You can do many things if you want. From simple to complex:

1.- Create your algorithm in Python or R and use it with the corresponding extension. In Python, you will need pandas to process both the input and the output from RapidMiner to your algorithm/script. I have done this for image transformation and it works.
2.- Create your algorithm in Java and use the Execute Script operator. I haven't done this.
3.- Create a library in Java with your algorithm, and implement your own operator as an extension. I am trying to finish one I have been writing since October last year (mostly because my other projects demand a lot of time).
4.- If the algorithm is in a language other than Python, R, Java or Groovy (e.g., I've done Ruby scripts for scoring stuff, and I still regret those), you may want to store information on a file and pass that file as a parameter to a Shell script that in turn returns you the output as a file that you can read. I've done this countless times before, please go with some of the other alternatives.

All the best,

Rodrigo.

rfuentealba

One more thing: By "create" I mean you can "use" it. You don't have to implement something from scratch if it's already in Python, for example.

divya_das

Thank you Rodrigo for the detailed explanation. 

Regards,
Divya