The number of links should not grow indefinitely.

LABSS / PROTON-T

Simulation of recruitment to terrorism

MIT License

1 stars 2 forks source link

The number of links should not grow indefinitely. #65

Closed mariopaolucci closed 5 years ago

mariopaolucci commented 5 years ago

This should be checked - when do they stabilize, and to what value?

digitaldust commented 5 years ago

it looks like links stabilize after 100 ticks.

Schermata 2019-07-11 alle 13 36 51

I wonder however if we should allow variability among agents, let's say that they can remember a random number of places up to the threshold in the GUI… this would also encourage routines, like people always going to the same 3 maybe 4 places… which sounds pretty realistic to me...

mariopaolucci commented 5 years ago

let's say that they can remember a random number of places up to the threshold in the GUI…

that would also amount to having (assuming uniform between 0 and value) to cutting the number of links in half.

Worth trying - you write it and see if it works?

digitaldust commented 5 years ago

only caveat, I'd do 1 + random links-cup, in order not to have people without any places

mariopaolucci commented 5 years ago

Maybe even more, as in 5 + random (2* links-cup - 5), also I'd suggest to make the slider have the obvious meaning of the mean

digitaldust commented 5 years ago

UPDATE: I am running two runs at 40k on my laptop, ~12 hours have passed and they are at ~ 350 ticks (on 600 total)… My feeling is that it's working, but tonight I should be able to compare results with the other runs (at 5k, 10k 20k, old 40k) and have a more clear picture. If it works, I guess computational time won't be a big issue anymore. I'll keep you posted here. NOTE: I am using 4GB for each run.

mariopaolucci commented 5 years ago

Cool. 4GB only? It crashed for Guy with 4...

I will have the 600-steps run on drake finishing soon, for comparison.

digitaldust commented 5 years ago

cool? I was aiming for awesome... ;-)

Yep it’s 4GB, don’t forget we don’t have 800k links anymore...

You ran from my branch on drake? Please send the csv as soon as you have it!

digitaldust commented 5 years ago

runs ended, it took 22,4 hours, see the plot, links are contained as expected:

I have to correct my statement about GB, I'm running with 2GB, here's my headless.sh: JVM_OPTS=(-Xmx2048m -XX:+UseParallelGC -Dfile.encoding=UTF-8)

no errors whatsoever in the log.

if we run the model with this adjustment, my prediction for 8640 ticks is ~14 days: (((80915.014+82847.074)/2)/60/60)*8640/600/24=13.646840666666662