tweedie

High performance Python GLMs with all the features!

Tweedie family density estimation in python

Midicoth - Micro-Diffusion Compression - Binary Tree Tweedie Denoising for Online Probability Estimation

[AAAI 2026, oral] unshrink: post-hoc debiasing of machine learning predictions to reduce shrinkage-induced attenuation bias in downstream causal inference -- without requiring fresh ground-truth data

Multi-source sparse Tweedie modelling

GLM regression (logistic, linear, Poisson, Gamma, Tweedie, negative binomial, multinomial) in pure DuckDB SQL — fit, predict & evaluate as table macros, with ridge, offset/exposure & weights

Conformal prediction intervals for Tweedie and Poisson insurance pricing models. Distribution-free coverage guarantees.

Frequency-severity GLM pricing framework in R on 678K French motor policies: Poisson frequency with exposure offset, Gamma severity, Tweedie benchmark, credibility-weighted rate table.

Spatially weighted conformal prediction intervals for geographically calibrated UK insurance pricing

End-to-end loss-cost regression and claim classification on 40K auto insurance policies. Tweedie XGBoost + isotonic-calibrated LightGBM. 1st place case competition.

An analysis of a Insurance Dataset with the goal of creating an ML model which can effectively predict insurance claims.

Sarmanov copula joint frequency-severity for insurance pricing — analytical premium correction, IFM estimation, dependency diagnostics

Zero-Inflated Tweedie Double GLM — three-head CatBoost EM (mu, phi, pi), balance check, Vuong test

Distributional GBM for insurance pricing — TweedieGBM, GammaGBM, ZIPGBM, NegBinomialGBM, per-risk volatility scoring

Actuarial pricing model for Brazilian auto insurance - GLM Poisson + Gamma, XGBoost Tweedie, SHAP, collision & theft separately, market premium analysis. SUSEP AUTOSEG 2019–2021.

Implementation of the paper : "Bayesian imaging using Plug & Play priors : when Langevin meets Tweedie"

Actuarial pure-premium pricing: Poisson x Gamma vs Tweedie, and why the sufficiency metric's own measurement noise means most model differences cannot be claimed

Conformal prediction intervals for Tweedie/Poisson insurance models - distribution-free coverage guarantees