CoarseFeatureSelectionOperator

#####
DebugHelper
Debug operator.
#####
HHHExtractionPlain
This operator calculates the set of Hierarchical Heavy Hitters for each Example in
 the given ExampleSet and returns an ExampleSet containing the sets of HHH. The HHH
 are stored as Objects in the ExampleSet via mapping: Each Example has
 an attribute that maps to an Object of Type <code>Set</code> containing 
 the HHH.
#####
HHHExtractionSpinczyk
This operator calculates the set of Hierarchical Heavy Hitters for each Example in
 the given ExampleSet and returns an ExampleSet containing the sets of HHH. The HHH
 are stored as Objects in the ExampleSet via mapping: Each Example has
 an attribute that maps to an Object of Type <code>Set</code> containing 
 the HHH.
#####
HistoExtraction
This operator calculates the histogram of system calls for each Example in
 the given ExampleSet and returns an ExampleSet containing these histograms.
#####
UnseenClassValidation
<p>
 This operator evaluates the performance of a feature selection
 algorithm. The evaluation is done in an unusual way: Normally, 
 we would evaluate the performance on unseen examples of the  
 classes used during feature selection. Here, we evaluate the performance 
 on unseen examples belonging 
 to previously unseen classes.</p>
 <p> 
 Consider the following example: We have log data of applications
 consisting of system calls (features). We want to predict the name
 of the application by looking at the log data. We also want to apply feature 
 selection, possibly eliminating irrelevant system calls.
 If the applications are fixed, we do the usual feature selection,
 determining which features to use on a sample containing the same 
 classes (applications) that are in the test set.
 However, we might want to know which features (system calls) are generally
 more relevant than others. This would allow us to make recommendations
 which features to use for learning the names of arbitrary, previously
 unseen applications. Obviously, thats a bit more difficult.  
 </p>
 <p>
 The first inner operator is the feature selection algorithm to be evaluated.
 Its input is an example set containing examples belonging to a randomly
 selected sample of classes. 
 It must return an attribute weights vector which is applied to the test data.
 </p>
 <p>
 The test data is an example set containing examples not belonging to the classes 
 selected for the first inner operator. 
 The second inner operator takes this test data and produces a performance vector.
 </p>
 <p>
 There are to ways to evaluate this performance. If the checkbox "auto_validation"
 is selected, the second operator is also run on a set of unseen examples 
 of the known classes (as in plain attribute selection),
 giving an estimate of the penalty incurred by selecting the attributes on the 
 wrong set of classes. The set of unseen examples of the known classes is produced 
 by splitting the set of examples of the known classes in two before selecting the attributes,
 so attribute selection is done on a set half the original size when this checkbox is selected.
 If the checkbox "all_attributes" is selected, the second operator is also run on the
 set of unseen examples with all attributes switched on.
 The second operator will thus be run up to three times for each subset of classes.
 It is guaranteed that these three runs are performed on example sets of equal
 size, so that none of the inner learning algorithms has an unfair advantage and
 the results are comparable.</p><p>
 The operator also provides extensive logging facilities.
#####


