all the text representations are generated regardless of set of features I need (e.g. I might not need POS tags but the input is POS-tagged anyway).
there is no information about which resources does a feature require.
I suggest:
writing a description for each feature stating which resources it needs
not extracting the information which is not needed for the set of features extracted (btw thus we can avoid the featureConfig variable. It now has 'bb', 'gb' and 'all' modes and extracts the information needed for all black-box features or all glass-box features or both). I don't know how exactly this is implemented in QuEst, but it shouldn't be difficult to change.
There are two sub-problems:
I suggest: