search

tnoulas / UrbanEpidemicSim

5 stars 2 forks source link

Transition probabilities #6

Closed kpelechrinis closed 4 years ago

kpelechrinis commented 4 years ago

I created transitions attribute for a Place object that is a dictionary with the transition probability from this place to others in the dataset. Currently it includes only transitions observed, but we should include a 'teleport'-like operation for venues that have not been visited from this place. This can either be part of the initialization phase or the actual simulation phase.

tnoulas commented 4 years ago

Thanks Kostas. When we add 'synthetic' residences I guess we can extend the 'matrix' of probabilities accordingly. Which make me think: how does time come into play in this? Do we want these transition probabilities to be sensitive to time? eg. I would think, in general, having say a different matrix for different periods of the day [morning, midday, evening, night] would make sense. Of course that would make the current matrix for sparse, but maybe a way around it would be to have transition probabilities between place categories and 'bias' at each period the matrix you created with the one that that you have in place. We could also use the Challenge dataset that has more transitions and in fact it has information on the period of the day that those happen. Happy to hear your views?

kpelechrinis commented 4 years ago

Yes I like both ideas. Having a different transition matrix for different times of the day and essentially having a two-step transition "decision"; first choose the next category based on a category transition matrix and then choose the restaurant based on a pruned version (considering categories and geography) of the venue transition matrix. Do you have quick access to the challenge dataset (NYC)? I have it on a server at school and I could not ssh for some reason ...