General structure test - big list of small array VS small list of big arrays.

[simoncarrignon]

So far the algorithm relies on a list of size N where each elements of the list is of size 7 and contain mix of tables/booleans/characters (in a full ABM way) . It would probably more efficient if the list was of size 7 and and each of these 7 elements would be array/matrix of size NxM, where m would depend on the size of each given properties. This would allow to compute all properties as vector in a much faster way ; but it means that each of the 7 NxM array will have to be increased at each birth (that's what I did in : https://framagit.org/sc/twitter-spread/ and : framagit.org/sc/pleistoclimate ). I have no idea what's the best trade-off and probably will simply try both option and see what's the fastest (fastest whcih will probably also depend on how fast pop are growing etc...)

[simoncarrignon]

maybe need to find way to implement social learning/transmission that is independent of the choice of this part of implementation ; and allow parallel works.

[simoncarrignon]

R/ modelVector.R : implement another version of the main structure of the model based on arrays. Obviously way faster:

The results are very similar, althought I found case where they were not ; but probably because of the order of marriage/reproduction/death being not the same in the vector vs list version (in the list version you can get married, give birth then die in a same year while your partner doesnt. This doesn´t happend with the array solution as everyone get updated at the same time

I still want need to explore a bit the parameters, It may happen than in some case the list would be better but so far it looks very unlikely wrt to the 20time spead increase. I will still keep the list version in a side, maybe useful later on.

[simoncarrignon]

Closing that, final decision/structure is arrays as much as possible!