ordinal-boost is the Python implementation of a novel mathematical framework for adapting machine learning (ML) models to handle ordinal response variables. Our approach is inspired by functional gradient descent used in gradient boosting and can be viewed as an extension of coordinate descent, where an ML regression model is embedded within a standard ordinal regression framework (e.g., logit or probit). The training process involves alternating between refining the ML model to predict a latent variable and adjusting the threshold vector applied to this latent variable to produce the ordinal response.
The workhorse is the BoostedOrdinal class, which conforms to scikit-learn conventions, including implementation of fit, predict and predict_proba methods. This allows for easy wrapping of the learner in hyperparameter tuning facilities provided in scikit-learn such as GridSearchCV.