germank / compatibility-naacl2015

Code for "So similar and yet incompatible: Toward the automated identification of semantically compatible words" in NAACL 2015 proceedings

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Compatibility Models

The description of these models will be published in the proceedings of NAACL 2015 under the title "So similar and yet incompatible: Toward the automated identification of semantically compatible words"

Requirements

Torch7

Usage

  1. Go to the directory model/torch and edit my_config.lua to point to the root of your local copy of this repository

  2. Create a "run configuration" (see the run_configs directory for examples)

  3. Train the model:

luajit do-train.lua run_configs/conf_01.lua
  1. Use the model to generate predictions:
luajit do-measure.lua run_configs/conf_01.lua

A pretrained configuration with the l2_interaction model is available at l2_interaction_config.lua

Troubleshooting

When I run the script, I get "luajit: do-train.lua:6: module 'models' not found:"

You need to add "./?/init.lua" to your LUA_PATH. One simple way of doing so is copy-pasting this command in your terminal:

`echo "export LUA_PATH=$(luajit -e 'print(package.path)');./?/init.lua"`

About

Code for "So similar and yet incompatible: Toward the automated identification of semantically compatible words" in NAACL 2015 proceedings

License:MIT License


Languages

Language:Lua 99.1%Language:Python 0.9%