Code for Pachitariu et al 2025
For Figures 2 and 3, the data will be shared upon publication of the paper (and is shared as a private, confidential link with the reviewers). The demo.ipynb notebook contains code to run an example simulation of linear dynamics with a random symmetric connectivity matrix. This notebook will take < 5 min to run on an A100 GPU. The expected output is included in the notebooks.
If you use any of this code, please cite the paper:
Marius Pachitariu, Lin Zhong, Alexa Gracias, Amanda Minisi, Crystall Lopez, Carsen Stringer. A critical initialization for biological neural networks. bioRxiv, 2025.
This code has been tested on Ubuntu 20.04 and Windows 11 with python=3.10, but python>=3.9 should work.
Install python, miniforge or anaconda recommended. Check the box to add anaconda to your path. Then open a new terminal or anaconda prompt (on Windows). Create a new environment and activate it:
conda create -n critical_init python=3.10
conda activate critical_init
Now clone the github repo and cd into it and install the requirements:
git clone https://github.com/mouseland/critical_init.git
cd critical_init
pip install -r requirements.txt
If you want to run jupyter notebooks in this environment, you will also need to run:
pip install notebook
If you want to use an Nvidia GPU to accelerate the code you need to first install Nvidia drivers on your machine. We have not tested the code with MPS from pytorch.
There are no strict version requirements, but listed below are the versions we used on Ubuntu 20.04:
- numpy=2.1.3
- rastermap=1.0
- scikit-learn=1.6.1
- torch=2.6.0
- torchaudio=2.6.0
- tqdm=4.67.1
- matplotlib=3.10.0
- natsort=8.4.0