Using TLO

These instructions are for using the Thanzi la Onse model model on the DIDE cluster with hipercow. We'll probably update some of the TLOmodel wiki once things stabilise and then we can perhaps point there as a source of truth.

The current state of play: We can currently run a simple copy of the model on the cluster, running on Python 3.11 under Windows. None of the Azure workflows are supported.

Minimal instructions, within a clone of TLOmodel.

First set up a hipercow root and configure it to use Python 3.11 on the DIDE Windows cluster:

hipercow init
hipercow driver configure dide-windows --python-version=3.11
hipercow driver show dide-windows

Install the dependencies and the model itself into a virtual environment on the cluster:

hipercow environment new
hipercow environment provision -- pip install --verbose -r requirements/dev.txt
hipercow environment provision -- pip install --verbose .

On the home drive, installation takes 200-300s for the development requirements and and 66s for the package itself. The requirements installation will appear to pause with no output for quite a while.

Optionally, confirm that things work:

hipercow task create --wait -- tlo --help

You can submit tasks with the tlo command to run a scenario. For example:

hipercow task create tlo scenario-run src/scripts/dev/scenarios/cluster-run.py

After which you can use hipercow task wait, hipercow task log and hipercow task status to keep an eye on your task. If you have many tasks to submit, you might need to use a loop in bash or similar.