catalyst-dynamic-text-classification

This code is a re-implementation of the paper Metric Learning for Dynamic Text Classification by ASAPP Research using Catalyst framework.

The original code for the paper can be found here asappresearch/dynamic-classification.

How to run

1. Clone repository

   git clone [email protected]:xelibrion/catalyst-dynamic-text-classification.git
   cd catalyst-dynamic-text-classification
   

2. Install dependencies

   pip install -e .
   

3. Fetch data

   cd dynamic_class
   ./get_data.py
   

4. Run train script to build vocabulary (it will fail to train the model without embeddings)

   ./train.py
   

5. Compute words vectors for the vocabulary using a fasttext model. Can be downloaded here.

   cat input/vocab.txt | awk -F ' ' '{print $1}' > vocab_words.txt
   ~/projects/fasttext/fasttext print-word-vectors  ~/projects/fasttext/cc.en.300.bin < vocab_words.txt > vocab_vectors.txt
   

   Please note that the original paper used GloVe as word embeddings. You might want to experiment with the choice of embeddings.

   Also, the tokenizer could be much better - at the moment it simply splits on whitespace.

6. Train the model

   ./train.py
   

Gotchas

This pipeline uses sru package, which might cause some challenges to get things running. See my comment here.