A batch-wise NLP tutorial with PyTorch

A simple batch-wise NLP tutorial using PyTorch.

Main requirements

• Python 3
• Pytorch
• conda

Installation

1. Clone this repo in your home directory (recommended)

   git clone https://github.com/epochx/pytorch-nlp-tutorial
   cd pytorch-nlp-tutorial

2. Create a conda environment. If you don't have conda installed, I recommend using miniconda. You can then easily create and activate a new conda environment with Python 3.6 by executing:

   conda create -n tutorial python=3.6
   conda activate tutorial
   

3. Run the installation script

   sh ./install.sh

Contents

sequence-classification

A tutorial for sentence classification using PyTorch. We start by showcasing the PyTorch workflow with a simple Logistic Regression for irony detection on the SemEval 2018 Dataset, using one-hot vectors.

Additionaly, we work on Sentiment Classification over the Large IMDB Movie Review Dataset, using a bidirectional LSTM with word embeddings.

Find the self-contained interactive Jupyter Notebook versions here:

• For the Logistic Regression:
  • https://goo.gl/cJn4qz
• For the LSTM:
  • https://goo.gl/LB9Z5C

To run the examples in your own machine:

• Download the data (will create ~/data/pytorch-nlp-tutorial):
  • sh ./sequence_classification/get_data.sh
• Run the Logistic Regression model:
  • python -m tutorial.sequence_classification.run_log_reg
• Run the LSTM model:
• python -m tutorial.sequence_classification.run_lstm

sequence-labeling

A tutorial for sequence labeling using a bidirectional LSTM and Conditonal Random Fields for NER for spanish over the CoNLL-2003 data.

Coming soon...