tf-glove

What is this?

This is an implementation of GloVe (Global Vectors for Word Representation), a model for learning vector representations of words. The model was originally developed by Jeffery Pennington, Richard Socher, and Christopher Manning.

This is my implementation of their model in TensorFlow, a "library for numerical computation using data flow graphs" by Google.

How do I use it?

Like this:

>>> import tf_glove
>>> model = tf_glove.GloVeModel(embedding_size=300, context_size=10)
>>> model.fit_to_corpus(corpus)
>>> model.train(num_epochs=100)
>>> model.embedding_for("reddit")
array([ 0.77469945,  0.06020461,
        0.37193006, -0.44537717,
        ...
        0.29987332, -0.12688215,], dtype=float32)
>>> model.generate_tsne()

For a more complete introduction, see the Getting Started notebook.

Credits

Naturally, most of the credit goes to Jeffery Pennington, Richard Socher, and Christopher Manning, who developed the model, published a paper about it, and released an implementation in C.

Thanks also to Jon Gauthier (@hans), who wrote a Python implementation of the model and a blog post describing that implementation, which were both very useful references as well.

References

• Pennington, J., Socher, R., & Manning, C. D. (2014). Glove: Global vectors for word representation. Proceedings of the Empiricial Methods in Natural Language Processing (EMNLP 2014), 12, 1532-1543.
• stanfordnlp/GloVe - Pennington, Socher, and Manning's C implementation of the model
• hans/glove.py - Jon Gauthier's Python implementation
• A GloVe implementation in Python - Jon's blog post on the topic