Sentimental

Sentimental is a project which allows to train a french sentimental analysis model based on CamemBert which is a french pre-trained transformer. The data used in this project is "Allociné" reviews so the model is well suited for movies sentiment analysis.

0- Clone the repository

git clone https://github.com/beroguedou/sentimental.git

1- Launch a training

Move in the cloned repository

cd sentimental

First of all you should create a directory with the name "saved" when the training function will save the future trained model.

mkdir saved

And then you could run a training by running the following bash command:

python train.py --epochs 30 --batch 3 --workers 5 --stopping 5 --device 'cuda:0' --name "sentimental_camembert_model.pth"

2- Serve a prediction

To serve prediction we made a basic flask application that is connected to a model trained (model should be in the directory "saved").

python sentimental_french_app.py