Frequently asked questions#

As you get started with emodpy, you may have questions. The most common questions are answered below. If you are using a disease-specific emodpy package, see the FAQs from that package for additional guidance. For questions related to functionality in related packages, see the following documentation:

Why does emodpy download a new Eradication binary each time I run?#

emodpy is designed to work much like a web browser: when you go to a website, the browser downloads html, png, and other files. If you visit the page again, it downloads them again so you always have the most current files. We want emodpy to work in much the same way. When you run simulations, emodpy will download the latest tested binary, schema, and supporting files that from the relevant EMOD ongoing branch.

However, if you need the stability of working from an older version, you can pass a Bamboo build number to emodpy.bamboo.get_model_files() to download that build instead. If you want to manually add a binary and and corresponding schema in the downloads directory to use, comment out the call to emodpy.bamboo.get_model_files() and nothing new will be downloaded.

What is the purpose of

The file contains all of your input and output paths in a single location. It also includes the path where model binaries (and associated schema) are downloaded to and uploaded from. Although you may ignore these files, it can be helpful to reference the schema for parameter information and have access to the binary itself.

I want to load a demographics.json file, not create one programmatically.#

Okay, but be aware that one of the benefits of emodpy and emod-api is that you get guaranteed consistency between demographics and configuration parameters to meet all interdependencies. However, if you want to use a raw demographics.json that you are very confident in, you can open that in your demographics builder. For example:

def build_demog():
    import emod_api.demographics.Demographics as Demographics
    demog = Demographics.from_file( "demographics.json" )
        return demog

What happens if I don’t connect to the VPN?#

You must be connected to the IDM VPN to access Bamboo and download the Eradication binaries (including plug-ins and schema). As an alternative, comment out the call to emodpy.bamboo.get_model_files() in the code and run the following (where “emod-disease” can be “emodpy-hiv”, “emodpy-malaria”, or “emod-measles”:

pip install emod-disease --upgrade
python -m emod-disease.bootstrap

The model files will be in a subdirectory called “stash.”

Why are the scripts read from the bottom?#

A Python script’s “main” block, which is also the entry point to the run script, appears at the end so that all the functions in the script have been parsed and are available. It is a common convention to structure the call flow bottom-up because of that.

My simulation failed on COMPS but I didn’t get an error until then#

The OS of the requested Bamboo build plan and the OS of the target platform need to match. For example, if your target platform is Calculon, the default, you’ll have to request a Linux build from Bamboo. There are no protections at this time (nor planned) to catch such misconfigurations.

How do I make my sim run inside a custom environment (on COMPS) for the first time?#

There are 3 small steps for this:

  1. Add a line of code:

    task.set_sif( manifest.sif )

    to your main Python script, after the task variable has been created.

  2. Add a line to your file like:

    sif = ""
  3. Create a new file called ‘’ – just match the name you used in step 2 – and put an asset collection id in it. At time of writing, this is the tested SIF asset id in the Calculon environment for running EMOD with Python3.9 and emod-api pre-installed:


You can find a quasi-catalog of available SIF ids here: InstituteforDiseaseModeling/singularity_image_files.

Note that you can of course just do this in one step, and add a line of code to your script like:

task.set_sif( "f1e6b032-47b7-ec11-a9f6-9440c9be2c51" )

But it’s much preferred to follow the above pattern so that future changes to use another SIF can be isolated to the resource file.

Is there a Singularity Image File that lets me run a version of the model that’s built against Python3.9?#

Yes. Assuming you already have a task.set_sif() call in your script, replace the current contents of your (or file with the following: f1e6b032-47b7-ec11-a9f6-9440c9be2c51. You may want to back up your existing version of that file.

What if I need a new or different SIF with a different custom environment?#

Anyone is free to create SIFs for themselves and use those. COMPS can build SIFs for you provided a ‘recipe’ – .def file. There are people at IDM who can do it on their desktops. Bear in mind Singularity really only installs on Linux.

How do I specify the number of cores to use on the cluster?#

num_cores is an undocumented kwargs argument to Platform. What that means is if you already have a script with a line like:

platform = Platform( "SLURM" )``

you would change it to something like:

platform = Platform( "SLURM", num_cores=4 )

to run with 4 cores.

What does “DTK” stand for?#

Disease Transmission Kernel. This was the early internal name of EMOD.

What is a “parameter sweep”?#

When the docs refer to a “parameter sweep”, it usually means an experiment consisting of a multiple simulations where almost all the input values are the same except for a single parameter. The parameter being swept will have different values across a range, possibly the min to the max, but any range of interest to the modeler. Parameter sweeps can be very useful for just learning the sensitivity of a given parameter, or as a form of manual calibration. A “1-D parameter sweep” is where you just sweep over a single parameter. You can also do “2-D parameter sweeps”, where you sweep over two parameters at once, and so on. But these of course require more simulations and fancier visualization.

A special kind of parameter sweep is sweeping over Run_Number which is the random number seed. This kind of sweep gives you a sense of the model to general stochasticity, given your other inputs.

You can sweep over config, demographics, or campaign parameters.

Is there any place where I can see which parameters are taken from distributions and what type of distributions are they?#

Any parameter that is being set from a distribution will have the distribution type in the name. E.g., Base_Infectivity_Gaussian_Mean tells you that this value is being drawn from a Gaussian distribution. If you don’t see any distribution name in the parameter name, it’s just fixed at that parameter value.