Optimization over several simulations

[leenr01]

Is there a way to perform an optimization on several simulations?
I defined a GUI experiment that creates three simulations each representing a city. I measure an average value across these cities that I want to minimize.
The problem is that any optimization is by itself an experiment (batch type) that creates its own simulations.
Also, unfortunately, GAMA doesn't seem to support nested experiments.

[benoitgaudou]

Dear, I admit it is not clear what you need here... Note that when you are using a batch experiment, you have the facet repeat that let you define the number of simulation for one parameter set... It seems close to what you are calling nested experiments, right ? cheers Benoit Le sam. 18 mars 2023 à 14:04, leenr01 ***@***.***> a écrit :

…

Is there a way to perform an optimization on several simulations? I defined a GUI experiment that creates three simulations each representing a city. I measure an average value across these cities that I want to minimize. The problem is that any optimization is by itself an experiment (batch type) that creates its own simulations. Also, unfortunately, GAMA doesn't seem to support nested experiments. — Reply to this email directly, view it on GitHub <https://github.com/gama-platform/gama/discussions/3676>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAL3WSWTP2ZH5N3MD5LXKN3W4WXFTANCNFSM6AAAAAAV7OYSWU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[leenr01]

Dear Benoit,

Thank you for your reply.
Not exactly.
Simply, I created several simulations that interact with each other using GUI experiment and I defined a value that is shared between them. Is there a way to optimize this value?

My model is quite long and complex. So, I created an analogy using model 9 in predator-prey. It is a simple model that creates three environments each with a different initial number of preys.
The code is pasted here. Can we maximize the variable "all_preys" using an optimization method inside GAMA?

---

model test2

global {
int simulation_ID <- 0;
map<int, int> nb_preys_init_list <- [1::500, 2::10, 3::3];
int nb_preys_init <- nb_preys_init_list at 1;
int nb_predators_init <- 20;
float prey_max_energy <- 1.0;
float prey_max_transfer <- 0.1;
float prey_energy_consum <- 0.05;
float predator_max_energy <- 1.0;
float predator_energy_transfer <- 0.5;
float predator_energy_consum <- 0.02;
float prey_proba_reproduce <- 0.01;
int prey_nb_max_offsprings <- 5;
float prey_energy_reproduce <- 0.5;
float predator_proba_reproduce <- 0.01;
int predator_nb_max_offsprings <- 3;
float predator_energy_reproduce <- 0.5;
int nb_preys -> {length(prey)};
int nb_predators -> {length(predator)};
int all_preys;

init {
	create prey number: nb_preys_init;
	create predator number: nb_predators_init;
}

reflex stop_simulation when: (nb_preys = 0) or (nb_predators = 0) {
	do pause;
} 

}

species generic_species {
float size <- 1.0;
rgb color;
float max_energy;
float max_transfer;
float energy_consum;
float proba_reproduce;
int nb_max_offsprings;
float energy_reproduce;
image_file my_icon;
vegetation_cell my_cell <- one_of(vegetation_cell);
float energy <- rnd(max_energy) update: energy - energy_consum max: max_energy;

init {
	location <- my_cell.location;
}

reflex basic_move {
	my_cell <- choose_cell();
	location <- my_cell.location;
}

reflex eat {
	energy <- energy + energy_from_eat();		
}

reflex die when: energy <= 0 {
	do die;
}

reflex reproduce when: (energy >= energy_reproduce) and (flip(proba_reproduce)) {
	int nb_offsprings <- rnd(1, nb_max_offsprings);
	create species(self) number: nb_offsprings {
		my_cell <- myself.my_cell;
		location <- my_cell.location;
		energy <- myself.energy / nb_offsprings;
	}

	energy <- energy / nb_offsprings;
}

float energy_from_eat {
	return 0.0;
}

vegetation_cell choose_cell {
	return nil;
}

aspect base {
	draw circle(size) color: color;
}

aspect icon {
	draw my_icon size: 2 * size;
}

aspect info {
	draw square(size) color: color;
	draw string(energy with_precision 2) size: 3 color: #black;
}

}

species prey parent: generic_species {
rgb color <- #blue;
float max_energy <- prey_max_energy;
float max_transfer <- prey_max_transfer;
float energy_consum <- prey_energy_consum;
float proba_reproduce <- prey_proba_reproduce;
int nb_max_offsprings <- prey_nb_max_offsprings;
float energy_reproduce <- prey_energy_reproduce;
image_file my_icon <- image_file("../includes/data/sheep.png");

float energy_from_eat {
	float energy_transfer <- 0.0;
	if(my_cell.food > 0) {
		energy_transfer <- min([max_transfer, my_cell.food]);
		my_cell.food <- my_cell.food - energy_transfer;
	} 			
	return energy_transfer;
}

vegetation_cell choose_cell {
	return (my_cell.neighbors2) with_max_of (each.food);
}

}

species predator parent: generic_species {
rgb color <- #red;
float max_energy <- predator_max_energy;
float energy_transfer <- predator_energy_transfer;
float energy_consum <- predator_energy_consum;
float proba_reproduce <- predator_proba_reproduce;
int nb_max_offsprings <- predator_nb_max_offsprings;
float energy_reproduce <- predator_energy_reproduce;
image_file my_icon <- image_file("../includes/data/wolf.png");

float energy_from_eat {
	list<prey> reachable_preys <- prey inside (my_cell);
	if(! empty(reachable_preys)) {
		ask one_of (reachable_preys) {
			do die;
		}
		return energy_transfer;
	}
	return 0.0;
}

vegetation_cell choose_cell {
	vegetation_cell my_cell_tmp <- shuffle(my_cell.neighbors2) first_with (!(empty(prey inside (each))));
	if my_cell_tmp != nil {
		return my_cell_tmp;
	} else {
		return one_of(my_cell.neighbors2);
	}
}

}

grid vegetation_cell width: 50 height: 50 neighbors: 4 {
float max_food <- 1.0;
float food_prod <- rnd(0.01);
float food <- rnd(1.0) max: max_food update: food + food_prod;
rgb color <- rgb(int(255 * (1 - food)), 255, int(255 * (1 - food))) update: rgb(int(255 * (1 - food)), 255, int(255 * (1 - food)));
list<vegetation_cell> neighbors2 <- (self neighbors_at 2);
}

experiment prey_predator type: gui {

output {
	display main_display {
		grid vegetation_cell border: #black;
		species prey aspect: icon;
		species predator aspect: icon;
	}	
}

init {
	loop i from: 1 to: 2 {
		create simulation with: [
			nb_preys_init :: nb_preys_init_list at (i+1),
			simulation_ID :: i
		];
	}
}

reflex global_nb_preys when: every(1#cycle){
	int sum <- 0;
	loop s over: simulations {
		sum <- sum + s.nb_preys;
	}
	all_preys <- sum;                // IS THERE A WAY TO MAXIMIZE all_preys in GAMA?
}

}

[benoitgaudou]

Dear, It seems indeed to be complicate here... But by the way, which parameters do you want to explore to maximize this value ? Other question : do you really want to maximize the number of prey for each step of the simulation? You want to find a set of parameters that makes all_preys at each step to each maximum value ? I admit that it would seem very surprising to me if you can find such a set. Cheers Benoit Le mar. 21 mars 2023 à 14:09, leenr01 ***@***.***> a écrit :

…

Dear Benoit, Thank you for your reply. Not exactly. Simply, I created several simulations that interact with each other using GUI experiment and I defined a value that is shared between them. Is there a way to optimize this value? My model is quite long and complex. So, I created an analogy using model 9 in predator-prey. It is a simple model that creates three environments each with a different initial number of preys. The code is pasted here. Can we maximize the variable "all_preys" using an optimization method inside GAMA? ------------------------------ model test2 global { int simulation_ID <- 0; map<int, int> nb_preys_init_list <- [1::500, 2::10, 3::3]; int nb_preys_init <- nb_preys_init_list at 1; int nb_predators_init <- 20; float prey_max_energy <- 1.0; float prey_max_transfer <- 0.1; float prey_energy_consum <- 0.05; float predator_max_energy <- 1.0; float predator_energy_transfer <- 0.5; float predator_energy_consum <- 0.02; float prey_proba_reproduce <- 0.01; int prey_nb_max_offsprings <- 5; float prey_energy_reproduce <- 0.5; float predator_proba_reproduce <- 0.01; int predator_nb_max_offsprings <- 3; float predator_energy_reproduce <- 0.5; int nb_preys -> {length(prey)}; int nb_predators -> {length(predator)}; int all_preys; init { create prey number: nb_preys_init; create predator number: nb_predators_init; } reflex stop_simulation when: (nb_preys = 0) or (nb_predators = 0) { do pause; } } species generic_species { float size <- 1.0; rgb color; float max_energy; float max_transfer; float energy_consum; float proba_reproduce; int nb_max_offsprings; float energy_reproduce; image_file my_icon; vegetation_cell my_cell <- one_of(vegetation_cell); float energy <- rnd(max_energy) update: energy - energy_consum max: max_energy; init { location <- my_cell.location; } reflex basic_move { my_cell <- choose_cell(); location <- my_cell.location; } reflex eat { energy <- energy + energy_from_eat(); } reflex die when: energy <= 0 { do die; } reflex reproduce when: (energy >= energy_reproduce) and (flip(proba_reproduce)) { int nb_offsprings <- rnd(1, nb_max_offsprings); create species(self) number: nb_offsprings { my_cell <- myself.my_cell; location <- my_cell.location; energy <- myself.energy / nb_offsprings; } energy <- energy / nb_offsprings; } float energy_from_eat { return 0.0; } vegetation_cell choose_cell { return nil; } aspect base { draw circle(size) color: color; } aspect icon { draw my_icon size: 2 * size; } aspect info { draw square(size) color: color; draw string(energy with_precision 2) size: 3 color: #black; } } species prey parent: generic_species { rgb color <- #blue; float max_energy <- prey_max_energy; float max_transfer <- prey_max_transfer; float energy_consum <- prey_energy_consum; float proba_reproduce <- prey_proba_reproduce; int nb_max_offsprings <- prey_nb_max_offsprings; float energy_reproduce <- prey_energy_reproduce; image_file my_icon <- image_file("../includes/data/sheep.png"); float energy_from_eat { float energy_transfer <- 0.0; if(my_cell.food > 0) { energy_transfer <- min([max_transfer, my_cell.food]); my_cell.food <- my_cell.food - energy_transfer; } return energy_transfer; } vegetation_cell choose_cell { return (my_cell.neighbors2) with_max_of (each.food); } } species predator parent: generic_species { rgb color <- #red; float max_energy <- predator_max_energy; float energy_transfer <- predator_energy_transfer; float energy_consum <- predator_energy_consum; float proba_reproduce <- predator_proba_reproduce; int nb_max_offsprings <- predator_nb_max_offsprings; float energy_reproduce <- predator_energy_reproduce; image_file my_icon <- image_file("../includes/data/wolf.png"); float energy_from_eat { list<prey> reachable_preys <- prey inside (my_cell); if(! empty(reachable_preys)) { ask one_of (reachable_preys) { do die; } return energy_transfer; } return 0.0; } vegetation_cell choose_cell { vegetation_cell my_cell_tmp <- shuffle(my_cell.neighbors2) first_with (!(empty(prey inside (each)))); if my_cell_tmp != nil { return my_cell_tmp; } else { return one_of(my_cell.neighbors2); } } } grid vegetation_cell width: 50 height: 50 neighbors: 4 { float max_food <- 1.0; float food_prod <- rnd(0.01); float food <- rnd(1.0) max: max_food update: food + food_prod; rgb color <- rgb(int(255 * (1 - food)), 255, int(255 * (1 - food))) update: rgb(int(255 * (1 - food)), 255, int(255 * (1 - food))); list<vegetation_cell> neighbors2 <- (self neighbors_at 2); } experiment prey_predator type: gui { output { display main_display { grid vegetation_cell border: #black; species prey aspect: icon; species predator aspect: icon; } } init { loop i from: 1 to: 2 { create simulation with: [ nb_preys_init :: nb_preys_init_list at (i+1), simulation_ID :: i ]; } } reflex global_nb_preys when: every(1#cycle){ int sum <- 0; loop s over: simulations { sum <- sum + s.nb_preys; } all_preys <- sum; // IS THERE A WAY TO MAXIMIZE all_preys in GAMA? } } — Reply to this email directly, view it on GitHub <https://github.com/gama-platform/gama/discussions/3676#discussioncomment-5380982>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAL3WSWSMFEFVHFJAVQQMGLW5GSCJANCNFSM6AAAAAAV7OYSWU> . You are receiving this because you commented.Message ID: ***@***.***>