I owed it to my tutor. And I’m not done. But it works at last (although, somewhat poorly for now).

My Project’s tutor published a paper some years back related to the “Mean Field” approach to a SIS infection simulation over network graphs.

In summary, the SIS simulator is approximated with a Markov Chain, i.e. only dependent on the last time step, using probabilities of infection of each node, their neighbours (through the adjacency matrix), and probabilities of cleaning up an infection for each node.

So instead of actually generating different configurations, and running different iterations whereby in each a lot of random numbers are generated (many times), we iterate until we reach a certain stability, directly multiplying probabilities.

And that’s a LOT fewer iterations.

And this entry is a note for the future dissertation/paper, marking the point in time at which I (finally) got into this topic. (I use these entries as timeline references of my progress so that the redaction of the final dissertation can be made somewhat easier thanks to referencing this blog… Hopefully, anyway).

As my tutor puts it, this new exercise makes the Project somewhat more complete, as we compare different methods and computational approaches.

Now I “just” need to find out, why my current implementation is working (the GA with MMCA does improve with iterations), but so much worse (in the first tests, at least) than the version which runs (many) more steps…

But well, the first implementation (functional) is there. And that’s quite good, as I have been delaying this exercise for way too long until now.