"Filtering the Results"

Question

Hey guys,

I want to compare the Rapid Miner for a university project with IBM Omnifind. For that I´d like to run the same scenario in both aplications. Don´t worry it´s a really simple one. I´ll give you the descripton and then what my problem is.

Scenario:
I use the NHTSA data base which contains many many problem reports of cars in America. I splitted every report in a seperate file. Now I want to compare the problem reports in a Correlation Matrix und filter it for the keyword fire. What I can see now is that I have a strong correlation between a car brand and a part of a car.

How to do this in RapidMiner:

I splitted the main file so that I have 1000 files each containing a problem report. Then I load the files via: 
Textinput->StringTokenizer->English Stopwordfilter->TokenLengthFilter->Porterstemmer.

After that I use the Correlation Matrix. The thing is that I get too many data. I want to filter the results so that I use only the files which contain the keyword I want to filter. In my case that is "fire". Is that possible? I get at the moment a wide range Correlation Matrix but can´t really use it. Plotting the results is not possible because of too much data.

I hope that you can help me.

Cheers
Benjamin

TobiasMalbrecht · Answer

Hi, you have to use the [tt]attribute_value_filter[/tt] option of the [tt]condition_class[/tt] parameter. As [tt]parameter_string[/tt] you have to specify a condition. Whenever an example does not fulfill the condition, it is filtered from the example set. The following code should work for your example. Hope that helps, Tobias

Legacy User · Answer

about the example filter. I set the parameter string to fire but I don´t really know how to set the condition class. Can you tell me what I need to set here. If I set the parameter string then I get from  every configuration that it doesn´t work with a parameter.

IngoRM · Answer

Hi,

the solution is quite simple: just use the operator "ExampleFilter" before applying the correlation matrix and filter out all examples where the TFIDF value for the keyword (here: Fire) or it's corresponding wordstem is 0. After that, you should apply a "RemoveUselessAttributes" operator to filter out all now constant attributes. Then apply the correlation matrix.

Cheers,
Ingo