intreg

intreg is a lightweight Python package for fitting interval regression models to censored and interval-bounded data using maximum likelihood estimation.

It provides:

• Interval regression (IntReg) for left-, right-, exact-, and interval-censored data
• Mixed-effects interval regression (MeIntReg) using random intercepts
• Optional L2 regularisation for fixed and random effects
• A clean NumPy/SciPy-based API suitable for modelling, research, and simulation

The package is useful for modelling censored continuous outcomes such as MIC values, detection limits, truncated biological measurements, and any interval-coded response.

---

📦 Installation

Install from PyPI

pip install intreg

Install from source (developer mode)

pip install -e .

---

🧩 Package Structure

The package exposes two core classes:

Class	Description
IntReg	Interval regression without random effects. Handles all types of censoring.
MeIntReg	Mixed-effects interval regression with random intercepts

Both models:

• use a Gaussian likelihood
• parameterise variance via log(sigma) for stability
• accept optional L2_penalties={} to regularise parameters
• optimise using scipy.optimize.minimize

Project layout (under src/):

intreg/
    __init__.py
    intreg.py
    meintreg.py

---

🔧 Basic Usage

1. Interval Regression (IntReg)

import numpy as np
from intreg import IntReg

y_lower = np.array([1.0, 2.0, -np.inf, 3.0])
y_upper = np.array([1.0, 2.5, 1.5, np.inf])

model = IntReg(y_lower, y_upper)
model.fit()

print(model.result.x)  # estimated [mu, log_sigma]

2. Mixed-Effects Interval Regression (MeIntReg)

import numpy as np
from intreg import MeIntReg

y_low = np.array([1.0, 2.0, 1.5, -np.inf])
y_high = np.array([1.0, 2.2, np.inf, 1.0])

X = np.column_stack([np.ones(4)])  # intercept-only model
groups = np.array([0, 0, 1, 1])

model = MeIntReg(y_low, y_high, X, random_effects=groups)
model.fit()

print(model.result.x)  # [beta, random_effects, log_sigma]

---

🧪 Testing

Tests are stored under:

temp/tests/

Run them using:

pytest

---

📄 License

MIT License.