/adj-relation

Primary LanguageTeX

Code base for the TACL Paper

Dependencies

  • python 2.7.13
  • yaml
  • pulp
  • numpy
  • sklearn

Instructions:

  1. Make sure all dependencies are installed
  2. Download the data for this experiment at [link/to/data] and place in directory path/to/data
  3. Git clone this repo to path/to/adj-relation
  4. Change the local field in adj-relation/config.yml file to path/to/data
  5. Open bash and set the path by:
export PYTHONPATH=/path/to/adj-relation
  1. Finally, cd into path/to/adj-relation. Note this repo contains all pre-run results in their respective experiments subdirectory (see below). To re-run experiments:
  • Run random baseline, run this command in bash:
python experiments/baseline/demos/uniform.py

You may find the results at adj-relation/experiments/baseline/results

  • Run non-random baseline for each data set:
python experiments/baseline/demos/ppdb_graph.py
python experiments/baseline/demos/ngram_graph.py
python experiments/baseline/demos/ppdb_ngram_graph.py

You may find the results at adj-relation/experiments/baseline/results

  • Run logistic regression for each data set:
python experiments/regression/demos/run_model_redo.py

You may find the results at adj-relation/experiments/regression/results

  • Run my replication of Mohit's MILP method:
python experiments/regression/demos/run_milp.py

You may find the results at adj-relation/experiments/milp/results

Warning: re-running these experiments takes a long time.

Contact.

lingxiao . at . seas . upenn . edu

Note

  1. Read methods.pdf for details of how the baseline and regression models are set up.

  2. This directory contains the scripts to re-run all experiments only. For scripts that collect the N-gram data, see https://github.com/lingxiao/learn-adj-relation. Warning: sparse documentation