- Prompt generated dataset of ~1000 samples that classifies monetary vs. non-monetary gifts.
- 6 models trained via cross validation over whole dataset, SVM with TFIDF vectorizer performed the best.
Download the project from Github and change your current directory:
$ (base) cd money_detect
Use a virtual environment to isolate your environment, and install the required dependencies.
$ (base) python3 -m venv venv
$ (base) source venv/bin/activate
$ (venv) pip3 install -r requirements.txt
To add a new model (add model with its vectorizer to models.py) or just train again with a new/updated dataset, write
$ (venv) python3 pipeline.py
The script automatically loops over all models, fits them to the dataset, and calculates confusion matrices, ROC/AUC curves and CV scores.
Model with the best CV score is saved with ONNX format as final_model.onnx.
When final_model.onnx is present, write
$ (venv) python3 load_use_model.py
And the inference will start for its input text.
