This repository contains implementations of ULMFit by Jeremy Howard and Sebastian Ruder applied to NLP tasks for the Spanish Language.
Francisco Ingham project for Classification of Spanish Tweets as of 2017 in the GeneralTASS Dataset was used as a starting point for hyperparameter tuning.
A F1(macro) score of 0.7308 is achieved(see notebook). The fine-tuning and training of the classifier should not take longer than a couple of minutes on a machine suitable for deep-learning with a very basic GPU.
git clone https://github.com/adai183/ULMFiT_es.git
cd ULMFiT_es
conda env create -f environment.yml
conda activate fastai_v07
jupyter-notebook