fastglmm 0.1.0

Fast Generalized Linear Mixed Models in Python

fastglmm

fastglmm is a fast implementation of a Generalized Linear Mixed Model (GLMM) for count/binomial data with variance components.
It is inspired by PQLseq (Sun et al. 2019; PMID: 30020412), with added flexibility and significant performance improvements.

✨ Features

• Supports Binomial family with logit link
• Order-of-magnitude faster than PQLseq
• Handles variance components (tau1, tau2) with options:
  • Fixed values
  • Inference from data
• Stable Newton-Raphson updates with adaptive step size
• Regularization for numerical stability
• Easy to use API similar to statsmodels

📦 Installation

You can easily install fastGLMM via Conda:

conda install -c conda-forge fastglmm

🚀 Usage

import numpy as np
from fastglmm import GLMM

# Simulated data
n = 100
np.random.seed(0)
X = np.hstack((np.ones((n, 1)), np.random.randn(n, 2)))
Y = np.hstack((np.random.randint(0, 10, (n, 1)), np.random.randint(1, 10, (n, 1))))
G = np.random.randn(n, 500)
K = G @ G.T

# Fit model
res = GLMM(X, Y, K).fit()

# Summary
param, coef = res.summary()
print(param)
print(coef)