https://github.com/ahmedbesbes/multi-label-sentiment-classifier

How to build a multi-label sentiment classifiers with Tez and PyTorch
https://github.com/ahmedbesbes/multi-label-sentiment-classifier

bert-model goemotions huggingface huggingface-transformers multi-label-classification nlp nlproc pytorch squeezebert tez

Last synced: 6 months ago
JSON representation

How to build a multi-label sentiment classifiers with Tez and PyTorch

• Host: GitHub
• URL: https://github.com/ahmedbesbes/multi-label-sentiment-classifier
• Owner: ahmedbesbes
• License: mit
• Created: 2021-02-23T18:26:51.000Z (over 4 years ago)
• Default Branch: main
• Last Pushed: 2021-02-28T18:28:53.000Z (over 4 years ago)
• Last Synced: 2024-05-08T00:26:05.808Z (about 1 year ago)
• Topics: bert-model, goemotions, huggingface, huggingface-transformers, multi-label-classification, nlp, nlproc, pytorch, squeezebert, tez
• Language: Jupyter Notebook
• Homepage:
• Size: 609 KB
• Stars: 18
• Watchers: 4
• Forks: 5
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
### Training a Multi-Label Emotion Classifier with Tez and PyTorch

If you're tired of rewriting the same boilerplate code of your training pipelines in PyTorch, I've found a pretty neat solution that could make your life easier. Don't worry, it's not a heavy library that'll change your way of doing things.

It's rather a lightweight wrapper that encapsulates your training logic in a single class. It's built on top of PyTorch, it's quite recent but I've tested it and I think it does what it promises so far.

It's called Tez and we'll see it today in action on a fun multi-label text classification problem. Let's jump right in.

![](./images/goemotions.png)

### Things that will be covered

- Using the Datasets library to load and manipulate go_emotions data

- Defining the training pipeline with Tez

- Training a SqueezeBert lightweight model for a multi-label classification problem and reaching +0.9 AUC on validation and test data

### Things that will be done next (PR are welcome)

- Deploying the model

- Crafting a small UI with React or Streamlit

### Link to the trained model

[Download it here](https://goemotions-with-tez.s3.eu-west-3.amazonaws.com/model.bin)