NLarge 1.0.0

Data augmentation for NLP

SC4001 NLarge

Purpose of Project

NLarge is a project focused on exploring and implementing various data augmentation techniques for Natural Language Processing (NLP) tasks. The primary goal is to enhance the diversity and robustness of training datasets, thereby improving the performance and generalization capabilities of NLP models. This project includes traditional data augmentation methods such as synonym replacement and random substitution, as well as advanced techniques using Large Language Models (LLMs).

Initializing Virtual Environment

We use Poetry in this project for dependency management. To get started, you will need to install Poetry.

pip install poetry

Afterwards, you can install the needed packages from Python with the help of Poetry using the command below:

poetry install

Repository Contents

• report.tex: The LaTeX document containing the detailed report of the project, including methodology, experiments, results, and analysis.
• example/: Contains example scripts for data augmentation and model training.
  • demo.ipynb
  • Result of test
• NLarge/: The main package containing the data augmentation and model implementation.
  • dataset_concat.py
  • llm.py
  • pipeline.py
  • random.py
  • synonym.py
  • utils/

Usage

To run the models and experiments, you can use the python notebooks in the example/ directory. The notebooks contain detailed explanations and code snippets for data augmentation and model training. For the results of the experiments, you can refer to the example/test/ directory.

We also refer the user to demo_attention.ipynb for a more detailed example of how to use the pipeline.py module. The notebook contains the code for training a model with attention mechanism using the NLarge library as a toolkit for data augmentation.

Compute limitation

Should you face computational limitation, you can use the datasets that we have preprocessed and saved in the example/llm-dataset/ directory. As the inference time for the Large Language Models (LLMs) can be quite long, we have preprocessed in advance such that end users can directly use the preprocessed datasets for training and testing purposes.

Development

While the library has been developed and tested, the library can be easily extended with additional data augmentation techniques or with new models to support the testing and research of the performance of different augmentation techniques.

The library can be easily extended with additional data augmentation techniques through creation of new modules or files in the NLarge package.

Website

You can access the PiPy page of the project from the link here: pypi page

Our github repository can be found here: github page

Contributing

Contributions to this project are welcome. If you have any suggestions or improvements, please create a pull request or open an issue.