TODO.txt

FEATURE      print ET_forest
                - statistiche
REFACTORING  mersenne by klib
FEATURE      "adaptive" feature subset selection
FEATURE      cross validation
                - shuffle
                - stratified
FEATURE      tools command line
                - ad hoc serialization
FEATURE      multi thread
                - generare in modo parallelo (vedi mersenne jump)
                - be careful with concurrent access to feature_deck
FEATURE      bias reduction via:
                - feature probability distribution estimation
                - outlier detection
FEATURE      train
                - modified gini diversity for classification
                - mad for regression
FEATURE      categorical features
OPTIM        eliminate recursion everywhere
OPTIM        compact memory via subproblem view
                - l2 cache optimization
                - subsample selection (e.g each tree fitted on random 
                                       10% of samples)
                idea: keep a vector of sample_idx pointed by the subproblem
REFACTORING  review api/headers etc.etc
DOC          include third party licenses
                - url: https://bitbucket.org/cmcqueen1975/simplerandom/wiki/Home
DOC          Api, examples

### recurring ###

REFACTORING  review double/float