The model runs with some sample data over a several years production. The model has 2 levels of hierarchy: 1. ‘farm’ level and ‘infection status of animals’ level. The spread between the farms is controlled by animal movements between the farms and the spread between animals inside the farms is controlled by 2 parameters in the model that are specific to FMD. As you might imagine, different diseases have different infectiousness and different progress on disease. Therefore the spread within-herd ie. between individual animals that have contact with each other is different for different diseases. However, the way disease spreads between herds if you exclude wierd things like spread through the air or via wildwild that may carry disease is simply that you move infected individuals from one herd to another and then start a spread process at the recipient herd. Within each herd you can think of the units there as animals, but the model doesn’t actually keep track of individuals, it only tracks how many individuals are in each of the infection states (SIR), this is known as a compartment model.
With the game in mind, we want a model that can be run somehow in the background and interacted with from whatever the game front end it. There are probably many ways to acheive this such as a live connection to an R process but a simple first approach might be to run the model in R for one day and get the result, use it for the game, potentially modify things and pass that data back to the model to run for another day. There would be several datasets that you would get at the end of a model day and can modify and pass for the next day.
- The state of each farm. This is a table that is one row per farm and one column per disease state. Each cell contains the number of animals in the disease state (compartment). In the SimInf framework we call this the “U” matrix. This is an integer matrix.
- The movement events of animals between farms, births, deaths and potentially between compartments. This is also a table and the strucuture is documented in the SimInf documentation. The structure is related to two matrices in the model that determines how these events are processed, the shift (N) and select (E) matricies.
- Other things: We can modify other things like the parameters of the model and if the model was more complex another matrix similar to U. This matrix is called V and is also one row per farm and any number of columns that can be continuous values.
Then a simple ‘game loop’ would be: Write out these datasets at the ‘end of a model-day’ to text files; Manipulate those files in some way in the game engine that the player interacts with; pass the text files back to Siminf and run another day and so on. I have written such an example in this repository and I would propose that we use some pre-generated events as the starting point for each day’s between herd spread and that if the player does nothing, then these are run; if the player intervenes then these are modified in some way.
There is another README in the game_model_example directory along with a prototype of how you might run the model from the game.