A project which solves SAT analogy questions using python and ConceptNet (http://conceptnet5.media.mit.edu/)
Python TeX CSS Go JavaScript
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
Report
submission
.gitattributes
SAT-package-V3.txt
TODO
conceptnetQuerier.py
conceptnetQuerier.pyc
correct.csv
data.csv
incorrect.csv
params.txt
querier.go
questions.txt
readme.md
satsolver.py
satsolver_metrics.py
set0.txt
set00.txt
set1.txt
set2.txt
set3.txt
submission.zip
termMap.pickle
termMap_backup.pickle

readme.md

ConceptNet SAT Solver

Running

To run the solver normally, do

./satsolver.py <question-file>

optional arguments:
	-v, --verbose  show every question, including relationships found and score
	-p, --pickle   pickle the term map after running (to cache the queries)

where <question-file> is one of the set_.txt files, for a small run of ten questions, or questions.txt for the full 374-question set.

WARNING: Running the full question set will take a long time if you don't have termMap.pickle in the same directory, likely somewhere between ten minutes and half an hour. However, this repo includes that cache, so it will probably take a minute or so.

The -p argument will cause the data fetched from ConceptNet to be written out to a file termMap.pickle. This file will then be loaded on every future run and used as a local cache in preference of requesting against the ConceptNet API. This will greatly speed future runs on the same test set.