Skip to content
Tensorflow implementation of Decomposable Attention Model
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

This is a quick and dirty Tensorflow implementation of an attention based NLP model that learns relationships between sentences. It is inspired by the paper: A Decomposable Attention Model for Natural Language Inference

The original paper has used the model on the Stanford Natural Language Inference (SNLI) dataset. I adapted this model to compete in Quora Question Pairs on Kaggle

There are three main files:

  1. defines a class that creates the model's graph. The paper has not described their model in complete detail. So it may differ in some ways, but I believe it captures the essence of what is described in the paper.

  2. loads and prepares training and validation datasets and iterates through question pairs one by one. It also saves model checkpoints in ./save directory, and produces a log on ./log for viewing in Tensorboard. No batching has been implemented.

  3. loads and prepares test data for the competition. It then restores the model from the latest checkpoint created by and iterates through the test data. Finally, it creates two CSV files ready to be submitted to Kaggle.

For a detailed description of this project and the results, please see this post.

You can’t perform that action at this time.