Process-ILP

ILP over semantic role classifier scores for collective role inference.

How to run?

$ python ilp_joint.py to run ILP optimization.
$ python evaluate.py to run evaluation module and get statistics.

Input/Output

• Cross-validation folds from Role Classifier is stored in data/cross-val directory. The input files are accessed from here.
• After ILP runs optimization over sentences of a process, output jsons are stored in the same input directory with the file name test.srlpredict.json.
• ILP assignments and objective functions generated for each process is saved in outputs directory.

Evaluation output

All evaluation outputs are stored in plots directory.

Config/Utils

• Use utils/ilp_config.py to configure the system. Please go through this file and set appropriate fields before running ILP/Evaluation.
• Most of the data processing happens in utils/ilp_utils.py
• Other utilities needed for entailment, memoization, word2vec scores, etc are in utils directory.

Code

Run pip install -r requirements.txt within ILP directory after cloning to install all the python dependencies.
The following modules are required to run the system:

• gurobipy
• NumPy
• Pandas
• pathos
• cPickle
• gensim
• retrying
• scikit-learn
• nltk
• Matplotlib
• Seaborn
• tqdm
• Requests

and textual entailment service by AI2 (Allen Institute for Artificial Intelligence).