The code is developed and tested on a machine with 8 NVIDIA GPUs with the CentOS Linux 7.4 (Core) operation system.


We do not recommend installation as root user on your system python. Please setup an Anaconda/Miniconda environment and add the required packages to the environment.

Please follow the steps below for a successful installation:

1 - Install PyTorch in a virtual environment

conda create -n py3_torch python=3.8
source activate py3_torch
conda install pytorch torchvision cudatoolkit=11.0 -c pytorch

More options to install PyTorch can be found here. Our package has been tested with CUDA 10.2 and 11.0. Then please ensure that at least PyTorch 1.8.0 is installed:

python -c 'import torch; print(torch.__version__)'
>>> 1.8.0

2 - Install PyTorch Connectomics

git clone
cd pytorch_connectomics
pip install --editable .

We install the package in editable mode by default so that there is no need to re-install it when making changes to the code.

3 - Ensure CUDA is setup correctly (optional)

Check that PyTorch is installed with CUDA support:

python -c 'import torch; print(torch.cuda.is_available())'
>>> True

Add CUDA to $PATH and $CPATH (note that your actual CUDA path may vary from /usr/local/cuda):

echo $PATH
>>> /usr/local/cuda/bin:...

echo $CPATH
>>> /usr/local/cuda/include:...

Verify that nvcc is accessible from terminal:

nvcc --version
>>> nvcc: NVIDIA (R) Cuda compiler driver
>>> Copyright (c) 2005-2020 NVIDIA Corporation
>>> Built on Wed_Jul_22_19:09:09_PDT_2020
>>> Cuda compilation tools, release 11.0, V11.0.221
>>> Build cuda_11.0_bu.TC445_37.28845127_0

Ensure that PyTorch and system CUDA versions match:

python -c 'import torch; print(torch.version.cuda)'
>>> 11.0

The codebase is mainly developed and tested on the Harvard FASRC cluster. For FASRC users, please load required CUDA modules from the RC server module list during running and development on the cluster. For example:

module load cuda cudnn


If you meet compilation errors, please open an issue and describe the steps to reproduce the errors. It is highly recommended to first play with the Jupyter notebooks to make sure that the installation is correct and also have an intial taste of the functions/modules.