order-embeddings-wordnet

Code for the hypernym completion experiment from the paper "Order-Embeddings of Images and Language". See the other repo for the caption-image ranking and textual entailment experiments.

Dependencies

Python 2 with a recent version of Numpy and nltk 3.0 for easy access to WordNet.
Torch7 with the argparse package.

Create Datasets

Run

python preprocessWordnet.py
th createDatasets.lua

Training the Model

To train with default hyperparameters (the order-embedding model from the paper), run

th main.lua --epochs 20 --name "myfirstmodel"

or train your own version by setting any of the the flags in main.lua.

The resulting weights are stored in weights.t7. You can view traces of training and validation error by navigating to the vis_training directory, running

python -m SimpleHTTPServer

and pointing your browser to the server (usually localhost:8000).

Reference

If you found this code useful, please cite the following paper:

Ivan Vendrov, Ryan Kiros, Sanja Fidler, Raquel Urtasun. "Order-Embeddings of Images and Language." arXiv preprint arXiv:1511.06361 (2015).

@article{vendrov2015order,
  title={Order-embeddings of images and language},
  author={Vendrov, Ivan and Kiros, Ryan and Fidler, Sanja and Urtasun, Raquel},
  journal={arXiv preprint arXiv:1511.06361},
  year={2015}
}

License