Pages: [1]
Author Topic: New Data Representation  (Read 908 times)
Posts: 38

« on: November 06, 2011, 01:06:23 AM »


I have a suggestion/comment on the future release  of  RM. As I remember you are working on new data structure/representation for RM6 I'd like to ask if you also plan to combine the new and the old structure data structure into some kind of common interface "DataSet". In my opinion the old ExampleSet has some advantages, that is efficiency for all learners that are based on distances/similarities or kernels. Of course it has some drawbacks like high memory usage when mixing different type of features (binary with numerical) but on the other hand this is efficient and fast!!!, that in my opinion is very important. The only thing that is missing at least for my problems are two methods in the Example and ExampleSet interfaces: the  object = getObjectValue(Attribute a) and void setValue(Attribute a, Object o) methods. That two methods would be very useful for any one who wants to use complex structure attributes like trees, graphs etc.
Of course if we have two different data representations, then each dataLoader should have an expert option to choose the internal data representation.

Pages: [1]
Jump to: