please I need help

boudi313

Hi  all

I have  questions in rapidminer :
1- I want to know if the logistic regression is the same of the maximum entropy or not if yes it is ok, if it no I want to know the operator name that represent the maximum entropy in rapidminer
2- I want to put 2 classifier one after one Respectively Like(logistic regression then KNN or any other classifier ) the aim of this is to improve the accuracy of text classification I want to know how to do this and how to measure the performance of this classification like in the picture


  3- Third one I want to know how to integrate my model in c# application and java application please I want in details the steps to do this and if there any video tutorial for this  .

Thank you
And I am sorry for this long post   

jcurry1

Hi Boudi,

Just one thing: It might be better to post these as separate questions because they are quite different.

Another thing: I am pretty new to this so don't treat my answers as being 100% correct but hopefully they might be helpful.

1.  Usually logistic regression has a binary outcome like yes/no or hit/miss.  It looks like you are talking about multinomial logistic regression where there are multiple outcomes or classes to predict e.g. low/medium/high. 

I just got that from http://en.wikipedia.org/wiki/Multinomial_logit
I don't know much more about it.

2.  I can't see the picture, but I imagine that normally you would do a classification of your data and then do another classification off of the data as another branch.  RapidMinre then has nodes that help you assess the performance of each classification.  It is probably worth going through the available videos to get an idea of the flow of a job in RapidMiner.

For text mining, I found the videos at http://vancouverdata.blogspot.ie/2010/11/text-analytics-with-rapidminer-loading.html very good.

3.  I don't know much about this yet.  I kind of presume there are nodes and things in RapidMiner that help in the exporting of your code.  RM produces java code under the hood so I imagine it is more straightforward to export to a  java application.  But I don't have experience of this yet.

MariusHelf

jcurry1 wrote:
3.  I don't know much about this yet.  I kind of presume there are nodes and things in RapidMiner that help in the exporting of your code.  RM produces java code under the hood so I imagine it is more straightforward to export to a  java application.  But I don't have experience of this yet.

That's wrong; RapidMiner does not produce Java code, but it is *written in* Java code. But it contains methods to load a model from XML or from the repository. So if you want to embed your models into another (java) application, you would use RapidMiner as a library and use it's methods to read your data and apply a previously generated model on it. There is whitepaper available in our shop on www.rapid-i.com which explains how to extend and/or use RapidMiner as a developer. Please keep in mind the licensing constraints from the fact that the RapidMiner Community Edition is released under the AGPL when you embed it into your own projects.

Converning the second question: please specify more detailed what extactly you are trying to do. Combine the classification results of two classifiers? Learn a new model on the predictions of another classifier? Both is possible; but as John said it is probably better to open a separate thread for that and describe your use case more detailed.

Best, Marius