parameters minFreqIndiv and nbStopWords

[erwanm]

issue migrated from original private gitlab repo

In the new implementation it is much more convenient to put all possible combinations of obs types together, e.g.

word.T.mf2.sw50
word.T.mf3.sw50
word.T.mf5.sw50
word.T.mf2.sw100
word.T.mf3.sw100
word.T.mf5.sw100

The disadvantage is the potential redundancy of information: if a basic obs type X, say POS trigrams, is very useful, it will be selected by the genetic algorithm multiple times with the different variants, possibly with little information gain between the variants.

In the original prototype the options were independent: there was only one possible value for each option in the config file, hence only one possible variant of each basic obs type selected. This can be implemented via some more complex config generation mechanism in the new version.

Since it is unclear to me at this point what the best solution is, the former option (easiest) is currently used.