Binary classification for sexist sentences

The final result of this project is a WebApp built with Angular and Python that allows you to classify a sentence, and help to train the dataset thanks to you feedback! Cool, right?

How to reproduce the experiment

To run the project you need

• Python3
• pip3

Once set up your Python environment you need to install some libraries like PyTorch, pandas ...

pip3 install pandas
pip3 install torch
pip3 install matplotlib
pip3 install seaborn
pip3 install scikit-learn
pip3 install transformers
pip3 install joblib

How to train the models

BERT

If you want to train BERT you need to run the sentiment_analysis_bert.py script. If you are on MacOS everything is fine, if you are on Linux/Windows you need to change the device used for training from mps to cuda

device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")

to

device = torch.device("cuda" if torch.backends.mps.is_available() else "cpu")

Naive Bayes

If you want to train the Naive Bayes you need to run the sentiment_analysis_naive_bayes.py script.

How to test the models

BERT

If you want to test BERT you need to run the test_bert.py script. If you are on MacOS everything is fine, if you are on Linux/Windows you need to change the device used for training from mps to cuda

device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")

to

device = torch.device("cuda" if torch.backends.mps.is_available() else "cpu")

Naive Bayes

If you want to test the Naive Bayes you need to run the test_naive.py script.