Dr Sarat Moka

PaRKol (Partial Rejection sampling for K-colouring) draws exact uniform samples of proper k-colourings of a graph. It implements the soft colouring framework, which decomposes the global sampling problem into small independent subproblems via partial rejection sampling, then solves each with coupling from the past. The algorithm is inherently parallelisable and achieves near-linear runtime.

COMBSS (continuous optimisation method toward best subset selection) reformulates the NP-hard discrete subset selection problem as a continuous optimisation over the hypercube, making it scalable to high-dimensional settings with p >> n. The package, jointly developed with Hua Yang Hu and Anant Mathur, now supports linear, logistic, and multinomial regression. Version 2.0.0 introduces a Frank-Wolfe homotopy method (Mathur et al., 2026) that selects a best subset of size k for each k = 1, …, q, with auto-calibrated penalties and optional LOOCV ridge tuning. The original Adam-based method for linear regression (Moka et al., 2024) is retained.

PyREGG estimates rare-event probabilities in Gilbert random geometric graphs using Naïve Monte Carlo, Conditional Monte Carlo, and Importance Sampling. It supports seven rare events including edge count, maximum degree, connected component size, triangle count, clique size, planarity, and forest. More rare events and higher-dimensional windows are coming soon.

This repository provides a Python implementation of our method Grid-FW proposed in Moka et al. (2025) for minimum-variance portfolio selection.

This repository provides Python code for reproducing the simulation results reported in Moka et al. (2025).