Using HNN in the cloud (no installation necessary)#

If you want to try out HNN without installing it locally on your computer first, there are several methods:

NSG cloud method#

HNN is installed for your use on the Neuroscience Gateway (NSG) Portal. To use HNN on this free resource, you should:

Login at https://nsgprod.sdsc.edu:8443/portal2/login!input.action or follow the link on that page to create a new account.
Follow the instructions at https://users.sdsc.edu/~kenneth/hnn/index.html. Note that you must right-click and save the provided files like benchmark_hnn_core_runtime.py directly from the webpage if you want to use them.

Please note that currently, HNN on NSG is using the previous version (v0.3), and only CORE functionality (not GUI) is supported. We will be updating NSG to use the most recent HNN version very soon.

Google CoLab cloud method#

You can click here to open our Google CoLab notebook which will run the HNN GUI in your browser. This is the simplest way to use HNN, but the simulations will take notably longer. This method does require a Google account, but the resources are free and allow you to run HNN GUI at no charge. To run the notebook, click “Runtime” then “Run all”, and follow the instructions carefully.

Installation Guide#

If you have any questions or problems while installing hnn_core, feel free to ask for help on our GitHub Discussions page!

Please follow the instructions in Steps 1, 2, and 3, in that order.

Step 1. Python Environment#

We strongly recommend that you install hnn_core inside a “virtual environment” using software like the Anaconda Python Distribution. If you are new to Python or data science in Python, we recommend you review the resources here: https://docs.anaconda.com/getting-started/.

Note that hnn-core currently only supports Python 3.8, 3.9, 3.10, 3.11, and 3.12, but not 3.13.

Step 2. Platform-specific requirements#

Windows#

Before you install hnn_core, it is important that you install the NEURON software system-wide using its Windows-specific binaries. You can test that NEURON was installed correctly by opening a Command Prompt (or similar) via your Anaconda install, and running the following command:
```
$ python -c "import neuron;"
```

MacOS#

If you already installed the NEURON software system-wide using its traditional installer package, it is recommended to remove it first. We will be installing NEURON using its PyPI package alongside hnn_core.
Before you install hnn_core, it is important that you install “Xcode Command-Line Tools”. This can be done easily by opening a Terminal and running the following command (followed by clicking through the prompts):
```
$ xcode-select --install
```

Linux#

If you already installed the NEURON software system-wide using its traditional installer package, it is recommended to remove it first. We will be installing NEURON using its PyPI package alongside hnn_core.

Step 3. `hnn_core` installation types#

Note that Step 1 and Step 2 above are required for all installation types. Note that you can install multiple sets of HNN features below during the install process by combining them similar to pip install "hnn_core[gui,parallel]".

Basic Installation#

To install just the hnn_core API, open a Terminal or Command Prompt using a fresh virtual environment, and enter:

$ pip install hnn_core

This will install only the hnn_core API along with its required dependencies, which include:

To check if everything worked fine, you can run:

$ python -c "import hnn_core"

and it should not give any error messages.

Basic Usage: For how to use the hnn_core API, which is available for every installation type, see our extensive Examples page.

Graphical User Interface (GUI) Installation#

To install hnn_core with both its API and GUI support, a simple tweak to the above command is needed (pay attention to the “quotes”):

$ pip install "hnn_core[gui]"

This will install hnn_core, its API dependencies, and the additional GUI dependencies, which include:

GUI Usage: To start the GUI, simply run:

$ hnn-gui

Optimization Installation#

If you would like to use Bayesian optimization, then scikit-learn is required. You can easily install hnn_core, its API dependencies, and its optimization dependencies using the following command (pay attention to the “quotes”):

$ pip install "hnn_core[opt]"

If you are planning to use Optimization, it is strongly recommended to also install support for Parallelism, which can be found below.

Optimization Usage: An example of how to use our Optimization support can be found in the Example here.

Parallelism#

hnn_core supports two kinds of parallelism:

Joblib parallelism: hnn_core can take advantage of the Joblib library in order to run multiple independent simulations simultaneously across multiple CPU processors (also called “embarrassingly parallel” jobs).
MPI parallelism: hnn_core can take advantage of NEURON’s use of the MPI protocol to split neurons across CPU processors, greatly reducing simulation time as more cores are used. This is more complex than other installation methods, but offers the best single-simulation speed. Note that we do not support MPI on Windows.

Parallelism: Joblib Installation#

You can easily install the hnn_core API, its API dependencies, and its additional Joblib parallelism dependencies using the following command (pay attention to the “quotes”):

$ pip install "hnn_core[parallel]"

This automatically installs the following dependencies:

Parallelism: Joblib Usage:

Once installed, you can run multiple trials simultaneously with only a small modification to your code, like this:

from hnn_core import JoblibBackend

# set n_jobs to the number of trials to run in parallel with
# Joblib (up to number of cores on system)
with JoblibBackend(n_jobs=2):
    dpls = simulate_dipole(net, n_trials=2)

Some in-depth examples of Joblib usage for trials are available in the Examples found here and also here.
Joblib can also be used to simultaneously run multiple simulations across “batches” of simulation parameters, illustrated in our Batch Simulation example.

Parallelism: MPI Installation#

For MPI installation, we strongly recommend you use the Anaconda Python Distribution specifically, since Anaconda is the easiest way to download OpenMPI binaries, which are not Python code.

Parallelism: MPI: Windows#

Unfortunately, we do not officially support MPI usage on Windows due to the complexity required. If this is a necessity for you and you do not have access to Linux/etc.-based HPC resources, please get in touch via our GitHub Discussions page. We may still be able to help.

Parallelism: MPI: MacOS#

First, create and activate your conda environment (but do not install hnn_core yet).
Inside your conda environment, install the conda-forge versions of OpenMPI and mpi4py using the following command:
```
$ conda install -y conda-forge::openmpi conda-forge::mpi4py
```

Next, copy, paste, and run the following commands to set some environment variables for your conda environment:

$ cd ${CONDA_PREFIX}
$ mkdir -p etc/conda/activate.d etc/conda/deactivate.d
$ echo "export OLD_DYLD_FALLBACK_LIBRARY_PATH=\$DYLD_FALLBACK_LIBRARY_PATH" >> etc/conda/activate.d/env_vars.sh
$ echo "export DYLD_FALLBACK_LIBRARY_PATH=\$DYLD_FALLBACK_LIBRARY_PATH:\${CONDA_PREFIX}/lib" >> etc/conda/activate.d/env_vars.sh
$ echo "export DYLD_FALLBACK_LIBRARY_PATH=\$OLD_DYLD_FALLBACK_LIBRARY_PATH" >> etc/conda/deactivate.d/env_vars.sh
$ echo "unset OLD_DYLD_FALLBACK_LIBRARY_PATH" >> etc/conda/deactivate.d/env_vars.sh
$ export OLD_DYLD_FALLBACK_LIBRARY_PATH=$DYLD_FALLBACK_LIBRARY_PATH
$ export DYLD_FALLBACK_LIBRARY_PATH=$DYLD_FALLBACK_LIBRARY_PATH:${CONDA_PREFIX}/lib

Next, install hnn_core with parallel support (pay attention to the “quotes”):
```
$ pip install "hnn_core[parallel]"
```

Finally, test that the install worked. Run the following command:

$ mpiexec -np 2 nrniv -mpi -python -c 'from neuron import h; from mpi4py import MPI; \
                                   print(f"Hello from proc {MPI.COMM_WORLD.Get_rank()}"); \
                                   h.quit()'

You should see output the looks like the following; this verifies that MPI, NEURON, and Python are all working together.

numprocs=2
NEURON -- VERSION 7.7.2 7.7 (2b7985ba) 2019-06-20
Duke, Yale, and the BlueBrain Project -- Copyright 1984-2018
See http://neuron.yale.edu/neuron/credits

Hello from proc 0
Hello from proc 1

Parallelism: MPI: Linux#

Note: these instructions are for installing hnn_core with MPI support on a personal Linux computer, not a High-Performance Computing (HPC) environment (also called a computing cluster or supercomputer). If you are on an HPC, then you should refer to your local HPC environment’s documentation on how to load the necessary MPI libraries/executables before installing hnn_core. Feel free to reach out to us via our GitHub Discussions page if you are unsure or run into problems; we should be able to help you get it working.

First, create and activate your conda environment (but do not install hnn_core yet).
Inside your conda environment, install the conda-forge versions of OpenMPI and mpi4py using the following command:
```
$ conda install -y conda-forge::openmpi conda-forge::mpi4py
```