An sklearn-compatible classifier for benchmarking NLP classification problems. The model used is the NBSVM described in section 2.3 of the paper Baselines and Bigrams: Simple, Good Sentiment and Topic Classification. The authors provide their own (matlab) implementation.
Simply clone the repo, cd into the project root directory and install into a python environment with pip install .
For example:
python3 -m venv venv
source venv/bin/activate
git clone git@github.com:fastforwardlabs/nbsvm.git
cd nbsvm
pip install -r requirements.txt
pip install .The NBSVM classifier is intended to be used on features transformed by either CountVectorizer or TfidfVectorizer.
Example usage looks like this:
from nbsvm import NBSVM
from sklearn.datasets import fetch_20newsgroups
from sklearn.feature_extraction.text import CountVectorizer
news = fetch_20newsgroups()
vectorizer = CountVectorizer(binary=True)
X = vectorizer.fit_transform(news.data)
y = news.target
model = NBSVM()
model.fit(X, y)
model.predict(X)There are a handful of unit tests for the public interface of the NBSVM class.
To run these locally, install the dependencies in requirements.txt into a clean environment and simply call pytest in the root directory of the project.
The first time the tests run, they will fetch a subset of the 20newsgroups dataset, which may take a few moments.
Tests should run in seconds after the initial download.
By default, the data will download to ~/scikit_learn_data (in your home directory), which can be changed by modifying the source.