This repository is used to make and understand concepts of NLP through multiple examples/models and data processing method.
All methods used will be written in a document when I'll consider to have got enough information to make a useful note. (This will likely be after the implementation of tranformers networks)
To clarify, my ressources to run the code come from the shadow computer most used for cloud gaming, here's the specs though:
- Intel Xeon CPU E5-2678
- 12 GB RAM
- Nvidia Quadro P5000
- main branch is used as a meddle of multiples models/methods
- tranformer branch is used exclusively for transformer network and to improve these
The main metric is F1 Score
Using a simple LSTM with only a coarse usage of "CountVectorizer": Train: ~ 90%+ | Eval: ~70% Using LSTM with Embedded layer and glove pretrained weights: Train: ~ 95%+ | Eval: ~40%
Might come back on these later (After transformer built)
WIP