Contributing to Hippunfold

Hippunfold python package dependencies are managed with Poetry, which you’ll need installed on your machine. You can find instructions on the poetry website.

HippUnfold also has a number of dependencies outside of python, including popular neuroimaging tools like wb_command, ANTs, c3d, and others listed in https://github.com/khanlab/autotop_deps. Thus we strongly recommend running HippUnfold with the --use-singularity flag, which will pull this container automatically and use it when required, unless you are comfortable using all of these tools yourself.

Note: These instructions are only recommended if you are making changes to HippUnfold code to commit back to this repository, or if you are using Snakemake’s cluster execution profiles. If not, it is easier to run HippUnfold when it is packaged into a single singularity container (e.g. docker://khanlab/hippunfold:latest).

Set-up your development environment:

Clone the repository and install dependencies and dev dependencies with poetry:

git clone http://github.com/khanlab/hippunfold
cd hippunfold
poetry install

Poetry will automatically create a virtual environment. To customize where these virtual environments are stored see poetry docs here

Then, you can run hippunfold with:

poetry run hippunfold

or you can activate a virtualenv shell and then run hippunfold directly:

poetry shell
hippunfold

You can exit the poetry shell with exit.

Running code format quality checking and fixing:

Hippunfold uses poethepoet as a task runner. You can see what commands are available by running:

poetry run poe

We use black and snakefmt to ensure formatting and style of python and Snakefiles is consistent. There are two task runners you can use to check and fix your code, and can be invoked with:

poetry run poe quality_check
poetry run poe quality_fix

Note that if you are in a poetry shell, you do not need to prepend poetry run to the command.

Dry-run testing your workflow:

Using Snakemake’s dry-run option (--dry-run/-n) is an easy way to verify any changes to the workflow are working correctly. The test_data folder contains a number of fake bids datasets (i.e. datasets with zero-sized files) that are useful for verifying different aspects of the workflow. These dry-run tests are part of the automated github actions that run for every commit.

You can use the hippunfold CLI to perform a dry-run of the workflow, e.g. here printing out every command as well:

hippunfold test_data/bids_singleT2w test_out participant --modality T2w --use-singularity -np

As a shortcut, you can also use snakemake instead of the hippunfold CLI, as the snakebids.yml config file is set-up by default to use this same test dataset, as long as you run snakemake from the hippunfold folder that contains the workflow folder:

cd hippunfold
snakemake -np

Instructions for Compute Canada

This section provides an example of how to set up a pip installed copy of HippUnfold on Compute Canada’s graham cluster.

Setting up a dev environment on graham:

Here are some instructions to get your python environment set-up on graham to run HippUnfold:

1. Create a virtualenv and activate it:

   mkdir $SCRATCH/hippdev
   cd $SCRATCH/hippdev
   module load python/3.8
   virtualenv venv
   source venv/bin/activate
   

2. Install HippUnfold

   git clone https://github.com/khanlab/hippunfold.git
   pip install hippunfold/
   

3. To run Hippunfold on Graham as a member of the Khan lab, please configure the neuroglia-helpers with the khanlab profile.

4. To avoid having to download trained models (see section below), you can set an environment variable in your bash profile (~/.bash_profile) with the location of the trained models. For Khan lab’s members, the following line must be add to the bash profile file:

    export HIPPUNFOLD_CACHE_DIR="/project/6050199/akhanf/opt/hippunfold_trained_models"
   

Note: make sure to reload your bash profile if needed (source ~./bash_profile).

5. For an easier execution in Graham, it’s recommended to also install cc-slurm snakemake profile for cluster execution with slurm.

Note if you want to run hippunfold with modifications to your cloned repository, you either need to pip install again, or run hippunfold the following, since an editable pip install is not allowed with pyproject:

    python <YOUR_HIPPUNFOLD_DIR>/hippunfold/run.py

Running hippunfold jobs on graham:

Note that this requires neuroglia-helpers for regularSubmit or regularInteractive wrappers, and the cc-slurm snakemake profile for cluster execution with slurm.

In an interactive job (for testing):

regularInteractive -n 8
hippunfold <PATH_TO_BIDS_DIR> <PATH_TO_OUTPUT_DIR> participant \
--participant_label 001 -j 8 --modality T1w --use-singularity \
--singularity-prefix $SNAKEMAKE_SINGULARITY_DIR

Where:

• --participant_label 001 is used to specify only one subject from a BIDS directory presumeably containing many subjects.

• -j 8 specifies the number of cores used

• --modality T1w is used to specify that a T1w dataset is being processed

• --singularity-prefix $SNAKEMAKE_SINGULARITY_DIR specifies the directory in which singularity images will be stored. The environment variable is created when installing neuroglia-helpers.

Submitting a job (for larger cores, more subjects), still single job, but snakemake will parallelize over the 32 cores:

regularSubmit -j Fat \
hippunfold PATH_TO_BIDS_DIR PATH_TO_OUTPUT_DIR participant  -j 32 \
--modality T1w --use-singularity --singularity-prefix $SNAKEMAKE_SINGULARITY_DIR

Scaling up to ~hundred subjects (needs cc-slurm snakemake profile installed), submits 1 16core job per subject:

hippunfold PATH_TO_BIDS_DIR PATH_TO_OUTPUT_DIR participant \
--modality T1w --use-singularity --singularity-prefix $SNAKEMAKE_SINGULARITY_DIR \
--profile cc-slurm

Scaling up to even more subjects (uses group-components to bundle multiple subjects in each job), 1 32core job for N subjects (e.g. 10):

hippunfold PATH_TO_BIDS_DIR PATH_TO_OUTPUT_DIR participant \
--modality T1w --use-singularity --singularity-prefix $SNAKEMAKE_SINGULARITY_DIR \
--profile cc-slurm --group-components subj=10

Running hippunfold jobs on the CBS server

1. Clone the repository and install dependencies and dev dependencies with poetry:

   git clone http://github.com/khanlab/hippunfold
   cd hippunfold
   poetry install
   

If poetry is not installed, please refer to the installation documentation. If the command poetry is not found, add the following line to your bashrc file located in your home directory (considering that the poetry binary is located under $HOME/.local/bin:

   export PATH=$PATH:$HOME/.local/bin

2. To avoid having to download containers and trained models (see section below), add the $SNAKEMAKE_SINGULARITY_DIR and $HIPPUNFOLD_CACHE_DIR environment variables to the bashrc file. For Khan lab’s members, add the following lines:

    export SNAKEMAKE_SINGULARITY_DIR="/cifs/khan/shared/containers/snakemake_containers"
    export HIPPUNFOLD_CACHE_DIR="/cifs/khan/shared/data/hippunfold_models"
   

3. HippUnfold might be executed using poetry run hippunfold <arguments> or through the poetry shell method. Refer to previous section for more information in regards to execution options.

4. On the CBS server you should always set your output folder to a path inside /localscratch, and not your home folder or a /srv or /cifs path, and copy the final results out after they have finished computing. Please be aware that the CBS server may not be the most efficient option for running a large number of subjects (since you are limited in processing cores vs a HPC cluster).

5. If you are using input files in your home directory (or in your graham mount in your home directory), you may also need to also add the following to your bashrc file (Note: this will become a default system-enabled option soon)

    export SINGULARITY_BINDPATH="/home/ROBARTS:/home/ROBARTS"
   

Deep learning nnU-net model files

The trained model files we use for hippunfold are large and thus are not included directly in this github repository, and instead are downloaded from Zenodo releases. If you are using the docker/singularity container, docker://khanlab/hippunfold, they are pre-downloaded there, in /opt/hippunfold_cache.

If you are not using this container, you will need to download the models before running hippunfold, by running:

hippunfold_download_models

This console script (installed when you install hippunfold) downloads all the models to a cache dir on your system, which on Linux is typically ~/.cache/hippunfold. To override this, you can set the HIPPUNFOLD_CACHE_DIR environment variable before running hippunfold_download_models and hippunfold.

Overriding Singularity cache directories

By default, singularity stores image caches in your home directory when you run singularity pull or singularity run. As described above, hippunfold also stores deep learning models in your home directory. If your home directory is full or otherwise inaccessible, you may want to change this with the following commands:

export SINGULARITY_CACHEDIR=/YOURDIR/.cache/singularity
export SINGULARITY_BINDPATH=/YOURDIR:/YOURDIR
export HIPPUNFOLD_CACHE_DIR=/YOURDIR/.cache/hippunfold/

If you are running hippunfold with the --use-singularity option, hippunfold will download the required singularity containers for rules that require it. These containers are placed in the .snakemake folder in your hippunfold output directory, but this can be overriden with the Snakemake option: --singularity-prefix DIRECTORY