Open chrishokamp opened 9 years ago
features can be computed 'online' -- i.e. at learning time, without a significant overhead. Representations take more time to compute, and may involve learning models over the whole dataset. This is a soft distinction, because some features take a long time to compute, and also require 3rd party resources, such as word net.
Representations add something to the contexts, they are a 'preprocessing' step for one or more feature extractors.
parsers take a filename and keys as input, read it, do some processing, and output an object. representations take data, filenames, and/or keys as inputs and output an object.
The output of representations and parsers is the same type, but their inputs are different.
To discuss: it's still not clear if there really is a difference between representations and parsers or not.