Parallel Classifier Combination in Single Process

kashif_khan

Hi ,

I am incorporating parallel classifier combination in rapidminer. I want to provide same vocabulary to all three classifiers in one go. I have  a seperate test set and know that i need to provide word list output of "Process Documents From File" (training) to "Process Documents From File" (testing). I am accessing my process via java code in which i created a process and access it via java code.

How can i programmatically change classifier in a process ?

kashif_khan

Its simple, Use "Multiply" operator which will create copies of input data. You can use those copies of data by proving it to different classifier.

kashif_khan

You can create a single process with multiple classifiers in it. All you have to do is to use two "Multiply" operators. 1 for the output example set of the "Process Documents From File" which is used for training set and 1 for "Process Documents From File" which is used for loading test set. This will create copies of the example set which you can apply to as many classifiers as you want...

MariusHelf

You will probably make your life easier if you prepare the process in the GUI, and use your program code just to exchange the classifier, or even to just execute the process

When including RapidMiner into your own software please keep in mind that RapidMiner is released under the AGPL, which means that also your code must be under the AGPL.

Best regards,
Marius

magedtab

Thank you for all

I have aproblem ,  i need to make combination of  knn and svm , i want to aggregate the output of them . i am waiting your reply .

keep in touch

MariusHelf

The Vote operator is your friend. Just put k-NN and SVM into its subprocess and you get a combined model.

Best regards,
Marius

magedtab

thank you Marius
i have another question.

what is the difference between vote operator, bagging and stacking ???

and what is your advice

keep in touch

MariusHelf

Vote is what you do - create different models with potentially different algorithms, and on application apply all and deliver the result of the majority.

Bagging means to create several, let's say 10, bootstrapped samples from the same dataset, train a model with the same algorithm on each, e.g. a decision tree. On application again perform a majority vote. The final bagging model is usually used to improve the stability and robustness of otherwise unstable methods like the decision tree. This is especially true on noisy training data.

Stacking finally means to create several models, and then train a final model on the predictions of the first group of models.

Best regards,
Marius

magedtab

thank you alot friend