This repository contains all code necessary to reproduce the figures in Condruz et al. (2026). The precomputed data required to run the figure notebooks is archived on Zenodo:
Download data.zip from Zenodo and unzip it in the root of this repository:
unzip data.zipBeyond the figure notebooks, this repository includes a standalone implementation of the PATNs (Periodic And Transient Networks) algorithm — a method for constructing recurrent neural network connectivity matrices with hand-crafted eigenspectra, Dale's law compliance, and controlled non-normality.
The algorithm lives in algorithm/ and comes with a self-contained manual. To get started, open PATNs_demo.ipynb or read PATNs_manual.pdf for a full description of parameters and usage.