• Besides the fact that LLMs have a huge power in generative use cases, there is a use case that is quite frequently overlooked by frameworks such as LangChain: Text Classification.
• 🚅 bullet was created to address this. It leverages the power of ChatGPT, while removing any boilerplate code that is needed for performing text classification using either Zero Shot or Few Shot Learning.

1. Install bullet: pip install "git+https://github.com/rafaelpierre/bullet.git#egg=bullet&subdirectory=src"
2. Configure your OPENAI_API_KEY
3. You should be good to go

from bullet.core.sentiment import SentimentClassifier


df_train_sample = df_train.sample(n = 50)

classifier = SentimentClassifier()
result = classifier.predict_pandas(df_train_sample)

# Define Few Shot examples

template = "Review: \"{review}\"\nLabel: \"{label}\""
examples = [
    template.format(
        review = row["text"],
        label = "POS" if row["label"] == 1 else "NEG"
    )
    for _, row
    in df_train.sample(3).iterrows()
]

df_test_sample = dataset["test"].to_pandas().sample(100)
reviews = df_test_sample.text.values

results = classifier.predict_few_shot(
    reviews = reviews,
    examples = examples
)

Full working example on a Jupyter Notebook can be found in notebooks/sandbox.ipynb

• From a terminal window, start a Python virtual environment and activate it:

python -m venv .venv
source .venv/bin/activate

• Install tox:

pip install tox

• Running unit tests:

tox

• From the docs folder, run pip install -r requirements.txt
• Run make html
• Documentation will be available at docs/_build/html/index.html