vivarium-cluster-tools 4.1.0

A set of tools for running simulation using vivarium on cluster.

Vivarium cluster tools is a python package that makes running vivarium simulations at scale on a Slurm cluster easy.

Supported Python versions: 3.10, 3.11, 3.12, 3.13

You can install vivarium-cluster-tools from PyPI with pip:

pip install vivarium-cluster-tools

or build it from source by cloning the monorepo and installing this package:

git clone https://github.com/ihmeuw/vivarium-suite.git
cd vivarium-suite
pip install libs/cluster-tools

Note

The [cluster] extra (i.e. pip install libs/cluster-tools[cluster]) pulls in jobmon_installer_ihme (hosted on IHME’s internal PyPI mirror), which is required to actually submit jobs on the IHME cluster. The base install (pip install vivarium-cluster-tools) resolves on public PyPI and is sufficient for read-only inspection of results / docs generation, but psimulate run etc. need the cluster extra.

A simple example

If you have a vivarium model specification file defining a particular model, you can use that along side a branches file to launch a run of many simulations at once with variations in the input data, random seed, or with different parameter settings.

psimulate run /path/to/model_specification.yaml /path/to/branches_file.yaml

The simplest branches file defines a count of input data draws and random seeds to launch.

input_draw_count: 25
random_seed_count: 10

This branches file defines a set of simulations for all combinations of 25 input draws and 10 random seeds and so would run, in total, 250 simulations.

You can also define a set of parameter variations to run your model over. Say your original model specification looked something like

plugins:
  optional: ...

components:
  vivarium_public_health:
    population:
      - BasePopulation()
    disease.models:
      - SIS('lower_respiratory_infections')
  my_lri_intervention:
    components:
      - GiveKidsVaccines()

configuration:
  population:
    population_size: 1000
    age_start: 0
    age_end: 5
  lri_vaccine:
    coverage: 0.2
    efficacy: 0.8

Defining a simple model of lower respiratory infections and a vaccine intervention. You could then write a branches file that varied over both input data draws and random seeds, but also over different levels of coverage and efficacy for the vaccine. That file would look like

input_draw_count: 25
random_seed_count: 10

branches:
  - lri_vaccine:
      coverage: [0.0, 0.2, 0.4, 0.8, 1.0]
      efficacy: [0.4, 0.6, 0.8]

The branches file would overwrite your original lri_vaccine configuration with each combination of coverage and efficacy in the branches file and launch a simulation. More, it would run each coverage-efficacy pair in the branches for each combination of input draw and random seed to produce 25 * 10 * 5 * 3 = 3750 unique simulations.

To read about more of the available features and get a better understanding of how to correctly write your own branches files,

Check out the docs!