All the files expect a folder called 'data' with the test data (1st subfolder should be duc2004) and a folder called 'rouge' with the rouge stuff in it (rouge pl script should be directly in there, as well as rouge's data folder). It also requires a folder called stanford-corenlp-full-2013-11-12 in the root directory, which you can get from here: http://nlp.stanford.edu/software/stanford-corenlp-full-2013-11-12.zip These are currently in the .gitignore, since they are very large.

sklearn pexpect unidecode xmltodict (can all be installed via pip)

The parsetest.py file shows the usage of the nlpio module. Only two functions are needed by a user, one being the loadDocumentsFromFile function, which loads all documents, including their manual models and peer suggestions, that are indexed by a file (here the testset.txt and the other function being the evaluateRouge function, where you put in the previously loaded documents as well as a list of your predictions (one per document) and receive a dictionary containing the recall, precision and F scores for the different rouge metrics along with their 95% confidence intervals. The test files also shows how to put in one of the peer suggestions as a prediction - maybe to establish a baseline. stanfordtest.py shows how to use the Stanford CoreNLP library to do parsing and POS tagging magic. Loading up this library can take a while, but then it's kept as a singleton.

The learntest.py file shows the usage of the nlplearn module. The modules contains transformers and estimators to set up an sklearn pipeline and do a parameter grid search through cross-validation. The branch basicsetup will always contain a 'clean slate' basic setup of the system with it always predicting the first sentence of a document as headline. Fixes and improvements to the general setup should be cherry-picked or merged into this branch.

• better file cleaning, maybe with some specialized tools

The corenlp package has been authored by Hiroyoshi Komatsu and Johannes Castner (https://bitbucket.org/torotoki/corenlp-python)