So yesterday for absolutely no reason, I had an idea to test an altogether different strategy.

My current project of SIS optimization with GA kinda works, yes. And I’m very happy about it. And I still need to try out the Markov-Chain approach (see if I can run things MUCH faster).

Anyhow, what if I were to choose iteratively the “best next node to act on to minimize infection prevalence”, instead of trying to find the best “overall budget distribution” at once? Would that work? Surely, it is only ONE approach, but it would also mean MUCH FASTER results, as I would only have to choose one node (among all) at each step and find which one gives the best result.

This means I would be simulating an increasing budget. First start with budget to act on one node only. Fix that when you find the best one. THEN suppose you are given ONE MORE budget point, so you get to choose how to use that ADDITIONAL improvement, on any remaining “node improvement”. This also means that if there were a better solution for choosing with TWO budget points upfront, I would almost certainly miss it. Which is why I call it a naive approach.

In other words, this approach CERTAINLY DOES NOT cover all possible distributions, and actually it covers much FEWER configurations than the GA approach (which itself was already only an approximation at best to the best solution). But does it even give “good solutions” at all, say compared to the GA?

I’ve already seen how page-rank is relevant to choose nodes to act on. I’ve already seen how generally speaking the GA chooses to act on more nodes instead of improving as much as possible fewer central nodes.

But I just I wonder what would happen with such a simplistic approach, in comparison. And so I just might try that “strategy” and compare it to the GA. Maybe I’ll do that this weekend. Or next week (more time and hopefully I’ll be a bit “less under the weather”).