This repo is my nlp pipeline that I am using for my NLP competitions on Kaggle.
Ideally, this will run similar to sklearn's fit_predict with a few edits on features that are generated.
It uses the following:
- wandb to record experimentation procedures
Currently supports the following competitions formats:
- Supports text multi-labelling
Built on:
- FastAI
- Pytorch
- transformers library for pre-built models
Assumes there is CUDA installed with Pytorch on computer.