Warning

This is unsupported legacy code

Readme

This project has a cabal and stack file. The easiest probably is to use stack.

The glpk library is required for building this project.

Run stack ghci to load all necessary files into a GHCi session.

Running it

In ghci use:

import PGF
import Data.Maybe
import qualified Subgrammar.GFRule as GFRule

positive_examples = ["I am here now"]
negative_examples = ["now I am here"]

rgl <- readPGF "/home/herb/src/gf/gf-rgl/src/english/Lang.pgf"
grammar = Grammar rgl ["/home/herb/src/gf/gf-rgl/src/english/LangEng.gf"]
positive_trees = examplesToForests grammar (fromJust $ readLanguage "LangEng") positive_examples
negative_trees = examplesToForests grammar (fromJust $ readLanguage "LangEng") negative_examples
problem = GFRule.forestsToProblem positive_trees negative_trees GFRule.numRules
solution <- solve problem
generateGrammar grammar (Just "/home/herb/src/gf/gf-rgl/src/") ["abstract","english","common","prelude","api"] solution True

Testing

There currently are two test functions Subgrammar.GFRule.test and Subgrammar.GFSubtree.

The test functions require the Exemplum grammar (both the PGF and the ExemplumEng.gf file) and the GF RGL

The relevant pathes can be adjusted in the Subgrammar/Common.hs file.

The tests load a grammar and use a set of example sentences to generate a new grammar. Afterwards it is tested that the examples are really covered.

Profiling

To profile the code you can build the code with profiling:

stack build --executable-profiling --ghc-options "-threaded"

or

stack install --executable-profiling --ghc-options "-threaded -rtsopts" --pedantic

Afterwards you can run it with profiling enabled:

stack exec --rts-options -p subgrammar-extraction

This results in a file called subgrammar-extraction.prof that can be analysed further.

By also adding -h you can also get the heap analysis as a .hp file that can be converted by using hp2ps.