Software on Hugh Zheng's website

latex2arxiv

hugh.stats@gmail.com (Hugh Zheng) — Wed, 20 May 2026 12:00:00 -0400

Open-source · github.com/YuZh98/latex2arxiv

Combinatorial Regression

hugh.stats@gmail.com (Hugh Zheng) — Wed, 20 May 2026 12:00:00 -0400

Open-source · github.com/YuZh98/combinatorial-regression

Reproducibility code for the paper Statistical Modeling of Combinatorial Response Data. Includes the MH-within-Gibbs sampler, kernel comparison experiments, and waterfowl matching data analysis.

Quick Start

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13


# Sanity check
make smoke
make probit_smoke

# Main simulation study (MH-within-Gibbs)
make full

# Kernel comparison (supplementary)
make kernel_compare

# Data analysis (waterfowl matching)
make duck_full # full model
make duck_reduced # reduced model

Simulation study

The default make full runs both exponential and half-Gaussian kernels. To run only the main-paper kernel:

combreg

hugh.stats@gmail.com (Hugh Zheng) — Sun, 19 Jul 2026 00:01:00 -0400

Open-source · github.com/YuZh98/combreg

combreg

Bayesian regression for combinatorial response data: each response is a binary vector constrained to an integral polytope, $\mathcal{Y} = {,y \in {0,1}^d : A y \le b,}$. Ignoring the constraint and fitting an unconstrained model leads to biased estimation, and enumerating all feasible outcomes is computationally impractical in high-dimensional settings. In Zheng, Ghosh & Duan (2026), an augmented likelihood that respects the combinatorial constraints is proposed, along with a Metropolis–Hastings-within-Gibbs (MH-Within-Gibbs) sampler that scales to high-dimensional problems.

Anti-correlation Gaussian Data Augmentation

hugh.stats@gmail.com (Hugh Zheng) — Wed, 20 May 2026 12:00:00 -0400

Open-source · github.com/YuZh98/Anti-correlation-Gaussian

Code accompanying the paper Gibbs Sampling using Anti-correlation Gaussian Data Augmentation, with Applications to L1-ball-type Models.

Files

File	Purpose
`SourceCode.R`	`l1ball.linreg(X, y, ...)`: blocked Gibbs sampler for the L1-ball-prior linear regression model using the anti-correlation Gaussian data-augmentation step
`SliceSampler.R`	Scalar and directional slice samplers (Neal, 2003). Used to update κ; can be sourced standalone
`LinearReg.R`	Toy variable-selection demo (`n = 300`, `p = 500`) producing estimation, trace, and ACF plots
`TruncatedMVN.R`	Extension of the anti-correlation trick to sampling a multivariate normal truncated to a box, plus a 2-D demo
`tests/test_samplers.R`	Smoke + correctness tests

Quick start

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12


source("SliceSampler.R")
source("SourceCode.R")

set.seed(1)
n <- 100; p <- 20
X <- matrix(rnorm(n * p), n, p)
beta_true <- c(rep(2, 5), rep(0, p - 5))
y <- as.numeric(X %*% beta_true + rnorm(n))

fit <- l1ball.linreg(X, y, steps = 4000, burnin = 2000, thin = 1,
 init_method = "random", verbose = FALSE)
round(fit$theta.est, 2)

To reproduce the paper’s linear-regression figures:

aldous-broder-lean

hugh.stats@gmail.com (Hugh Zheng) — Mon, 26 Oct 2026 13:00:00 -0400

Open-source · github.com/YuZh98/aldous-broder-lean

Machine-checked proofs of the main results of my paper The Order Mass Identity of Rooted Forests, with a Direct Proof of the Extended Aldous–Broder Theorem (arXiv:2610.13063, 2026). To my knowledge, this is the first machine-checked proof of the Aldous–Broder theorem.

latex2ufdissertation

hugh.stats@gmail.com (Hugh Zheng) — Wed, 20 May 2026 12:00:00 -0400

Open-source · github.com/YuZh98/latex2ufdissertation

A safety-net validator for UF doctoral dissertations using the Fall 2025+ University of Florida LaTeX template. Given a project archive, project directory, git URL, or compiled PDF, it produces a severity-tiered report citing the originating UF rule for each finding — one more pair of eyes before clicking submit.

NoteBuilder

hugh.stats@gmail.com (Hugh Zheng) — Mon, 26 Oct 2026 12:00:00 -0400

Open-source · github.com/YuZh98/notebuilder

Study material that does something. Build notebooks from reactive cards, drill them in Study mode, and hand them to anyone as a single offline file.

RecVAE — Recurrent 3D-Conv VAE for fMRI

hugh.stats@gmail.com (Hugh Zheng) — Wed, 20 May 2026 12:00:00 -0400

Open-source · github.com/YuZh98/VAE-fMRI-Alzheimer

A recurrent 3D-conv VAE for resting-state fMRI volumes, in the same family as Kim et al. (2021) but with a linear latent transition solved in closed form by ridge regression rather than learned by SGD. Ships with a synthetic data path so you can run it without ADNI access; every example fits on a laptop CPU.

Academic Application Tracker

hugh.stats@gmail.com (Hugh Zheng) — Wed, 20 May 2026 12:00:00 -0400

Open-source · github.com/YuZh98/academic-application-tracker

A local Streamlit dashboard that answers one question every morning: “What do I do today?”

Track dozens of postdoc, PhD, faculty, and fellowship applications in parallel — deadlines, recommendation letters, materials checklists, interview rounds — without a single missed follow-up.

Quant Prep Terminal

hugh.stats@gmail.com (Hugh Zheng) — Tue, 02 Jun 2026 12:00:00 -0400

Open-source · github.com/YuZh98/quant-prep-terminal

Everything you need to prepare for a quantitative-finance interview, packed into one HTML file: cheatsheets, flashcards, and a small lab where you can build an option position and watch its payoff in real time. No install, no sign-up — open it and study.