introduction.qmd

---
title: "Welcome and opening remarks"
format:
  revealjs:
    logo: "images/logo.png"
    slide-number: c/t
    preview-links: auto
    footer: "[mrc-ide/odin-monty-workshop-2025](.)"
execute:
  echo: true
  message: true
  output: true
  warning: true
---

# Safety first

## 🔥🔥🔥 In case of fire 🔥🔥🔥

::: {style="text-align:center"}
![](images/Evacuation.png){width=90%}
:::

## 🛜 WIFI

- EDUROAM
- for non-Academic participants

**The Cloud**

1. Connect to '_The Cloud' on your device
2. Follow the instructions

## Support

![](images/HPRU MRC Imperial joint.png)

# A brief history of odin

## 🏺 "Classic" odin - the beginnings (2016-2019)

- **`odin`** created to integrate ODEs (e.g. for compartmental models) in R with a domain-specific language (DSL)
- Limited support for difference (discrete-time) equations
- Automatic translation to C; efficient solutions in little code
- Used at Imperial for malaria, HIV, ebola and other diseases
- No support for inference

## 😷 COVID-19 response (2020-2022)

::: {style="text-align:center"}
![(Knock et al 2021, Science Translational Medicine)](images/knock.jpeg){width=60%}
:::

## 😷 COVID-19 response (2020-2022)

- **`mcstate`** for statistical machinery (particle filter, pMCMC)
- **`dust`** for efficient parallel simulation
- **`odin.dust`** compile `odin` code to use `dust`
- Collaborative work by the UK real-time modelling & research software engineers teams at Imperial College
- Many, many, rough edges

## Meanwhile...

* Different versions of the stack have been picked up outside of Imperial
  - Zurich, Munster, LSHTM, CDC, Fudan, Lancaster, Pasteur, NC State, Norway, Switzerland
* People tried to repurpose statistical machinery in `mcstate`
* We hit limits of computational efficiency and ability to manage inputs and outputs with the COVID model
* The documentation (22 vignettes) and packages (>5) were hard to navigate and discover

# Version 2

## New software

- Design of a new architecture, rewiring data, model and parameters
- New statistical interface, **`monty`**
  - A new small BUGS-inspired DSL for priors
  - Works well with `odin` models, but usable on its own
  - Modular, and eventually easy to extend
  - Fully replaces `mcstate`

## New community

- We want to understand how people are using these tools
- What have you built, what do you want to build?

## 🙌 For us to know

Who identifies as

1. 🟢 **Novices** (no prior tool experience)
2. 🟡 **Users of other tools** (e.g. Stan, JAGS, BayesTools)
3. 🔵 **odin/mcstate/monty users**

## Aim of the workshop

1. Introducing the 2nd generation of the toolkit
2. Collecting feedback
3. Building a community and fostering collaboration

## Overview of workshop

Day 1: Introducing the Tools and Applications

Day 2: Feedback, Development, and Support

📸 Group photo 1:40PM

🥂 Reception 5:30PM

🍽️ Workshop Dinner @ "The BroadCaster" 7PM