This repository contains the source code and data used for the talk Solving Natural Language problems with scarce data presented at the Big Things conference in 2019.

To reproduce the experiments presented in the talk you will need to install an Anaconda Python 3 distribution, clone this repository, then create a Conda environment with

conda env create -f environment.yml

You can then login into the environment with

source activate bigthings

The project is composed as a set of notebooks that approach the problem using different techniques. You can open Jupyter Notebooks to explore them by running

jupyter notebook

Index of notebooks:

• sklearn-baselines.ipynb: baseline models created with scikit-learn.
• keras.ipynb: deep learning models created with Keras.
• simpleBERTusage.ipynb: simple examples showing the usage of the BERT tokenizer and embeddings.
• bert.ipynb: BERT model created with Transformers.
• plotting.ipynb: results visualization.

You will also find a toxic.py file with auxiliary functions to load and preprocess the data.

Learn and enjoy!