LSS Structure

[schalkdaniel]

To boost LSS models the following things needs to be done:

• the base-learner track should be able to get initialized with a multidimensional response, then a hash map is created with vectors for each response
• The optimizer needs to loop over the map (controlled from compboost.cpp)
• The loop (in compboost.cpp) should update the response directly, then the next "response model" could use the updated response (corresponds to cyclic updates)
• Think about how estimated parameter are returned, how selected base-learners are returned (maybe as matrix) and so on
• Think about how stopping criteria comes in here
• We need special loses for that structure, the loss should control how the response looks like (maybe)
• Each response should get an optimizer (just to provide modularity)