I am somewhat new to Netlogo, and have been scratching my head over some of the results I get from behaviorspace. I have been playing with the wolf-sheep predation model, and changing their movement based on what color patch they are in. There is a single wolf and a single sheep, and what I want to measure is the number of time steps it takes for the wolf to eat the sheep. The patches are colored randomly, based on some proportion from 0-100, as below:
to color-patches
let total 100
let p-red slider1 / total
let p-green total - p-red
ask patches [
let x random-float 1.0
if x <= p-red + p-green [set pcolor green]
if x <= p-red [set pcolor red]
]
end
The issue I am having is when I set the movement of the wolf and the sheep to move independently of the patch color:
ask sheep [
set heading random 360
fd 1
]
ask wolves[
set heading random 360
fd 1
eat-sheep
]
My expectation is that the mean and standard error for the number of time steps until the wolf eats the sheep should be pretty similar regardless of how many red patches and how many green patches there are, since their movement is not affected by it. I ran it in behaviorspace, with 1000 iterations per 10% increase in proportion of red patches (from 0 - 100%). However, I keep getting results that look kind of like this:
enter image description here
Basically the means+se are all over the place. Every time I run it, they are distributed differently (but same grand mean). This is particularly odd since when I introduce any sort of patchcolor-specific behavior for either the wolf or the sheep, I get very clear patterns with less variation.
Any ideas what might be going on here? The only thing I could think of is that relative starting position of each is pretty important (but each is placed at random xy coords). I assumed that in behaviorspace for each iteration for a given set of parameters, it would run through all the code (thus generating a new random landscape and new random starting points for the wolf and the sheep for each of the 1000 runs per parameter combination). Does behaviorspace maybe take the first landscape and starting coordinates for each turtle and use them for each of the 1000 iterations per parameter combination?
Thanks!
Now I think you may not have a mistake but instead are just misinterpreting your graph. The Y axis on the graph you posted ranges only from 2000 to 2200; if you set the Y axis scale to 0 to 2500, the results from each experiment would look very similar to each other.
The difference in mean between your results (~2100) and my results (~3100) is probably just due to different world sizes. I presented standard deviation while you graphed standard error.
If you histogram the results, they seem to follow an exponential distribution.
Related
I have each patch holding two lists: directions and magnitudes. Both lists contain values that are computed based on nearby turtles:
directions contains the headings from the patch to the various turtles (so basically it is just the turtle's heading rotated by 180°);
magnitudes contains a value that is directly proportional to the turtles' mass and inversely proportional to the distance between the patch and the turtles;
The items of the two lists are coupled in their order, i.e. the first item of directions and the first item of magnitudes are computed based on the same turtle, the two second items are computed based on another turtle and so on.
What I want to achieve is to come up with two single patch-own values, my-direction and my-magnitude, representing in some way the weighted average of directions, where the weights are magnitudes.
To put it another way, I am thinking of this in terms of vectors: turtles are exerting on the patch a force that can be represented as a vector, always pointing in the direction of the turtle and with a certain intensity (the magnitude). The resulting vector (represented by my-direction and my-magnitude) should be the resulting average of these forces.
I have seen this question. It does not address the issue of a weighted average; however it mentions the concept of circular mean. I've delved into it a bit, but I'm not sure how to apply it to my case and even if to apply it: does it still apply even with the formulation of the problem in terms of vectors?
I've seen this question/answer on SE Mathematics where it is said that the average vector can be found by averaging x- and y-coordinates of the initial vectors. In my case, ideally all the pairs of values in the two lists form a different vector with origin in the patch at issue, with heading found in directions and length found in magnitude. I suspect I can find the coordinates of each vector by multiplying its sine and cosine by its magnitude, however at this point I'd use some guidance as I might be overcomplicating things (either from the maths perspective or the NetLogo perspective).
This one below is a reduced version of the code that brings to the point where target-patches (not focusing on all patches, in order to make it quicker) have the two lists populated.
globals [
target-patches
]
turtles-own [
mass
reach
]
patches-own [
my-direction
my-magnitude
directions-list
magnitudes-list
]
to setup
clear-all
set target-patches n-of 10 patches
ask target-patches [
set directions-list (list)
set magnitudes-list (list)
set pcolor yellow + 3
]
create-turtles 10 [
move-to one-of patches with [(not any? turtles-here) AND (not member? self target-patches)]
set mass (random 11) + 5
set reach (mass * 0.8)
set size (mass / 8)
set shape "circle"
]
populate-lists
end
to populate-lists
ask turtles [
let relevant-targets (target-patches in-radius reach)
ask relevant-targets [
set directions-list lput (towards myself) (directions-list)
set magnitudes-list lput (magnitude-based-on-distance) (magnitudes-list)
]
]
end
to-report magnitude-based-on-distance
report [mass] of myself / (distance myself * 1.2)
end
Your initial instinct is right. You can do something like
LET my-dx mean (map direction magnitude [ theta scalar -> scalar * sin theta ])
(I may have that map and anonymous syntax wrong please edit)
And do the same for my-dy using cos (edit: or maybe negative cos?)
patch-at my-dx my-dy is one way of getting the patch.
You can also do (atan my-dx my-dy) to gets the new direction
And distancexy my-dx my-dy to get the magnitude
Don’t remember if patch can use distancexy or patch-at, but hopefully so. Otherwise you have to use a helper turtle, and do the math yourself.
I think I did a orbital mechanics toy model once that did something like this. It’s hiding on turtlezero.com.
I have the set of points implemented in netlogo and agents are moving from one point to another. Each point has a weight (number approximately between 0 and 9, its not a probability). What I want to made is a simple rule.
I want to give all points probability of visit by the value of weight.
So the next point which will be visited by agent should be calculated by the probability based on point weight and the closeness point (more close point - bigger probability), but that closeness isn't so much big factor as the point weight. For example, I would like to set in formula that closeness is twice lower factor then point weight.
I investigated rnd extension, but I am not sure how to append probabilities to points which I am having a lot (approximately around 250 points).
You're on the right track with the rnd extension. From that extension you need the weighted-one-of primitive and you just put the formula into the reporter block.
I think this is something like what you want. It's a complete model so you can run it and see what it does. The reporter block uses the weight and the distance in the probability. Since you want the probability to be larger for closer, then I have used the inverse of the distance, but you could simply subtract the distance from something like the maximum distance in the model. You will also need an appropriate scaling factor (replacing the 10 in my example) so that the weight is worth twice an average value of closeness.
extensions [rnd]
turtles-own [weight]
to testme
clear-all
create-turtles 10
[ setxy random-xcor random-ycor
set weight 1 + random 3
set size weight
set color blue
]
ask one-of turtles
[ set color red
let target rnd:weighted-one-of other turtles [ 2 * weight + 10 / distance myself ]
ask target [ set color yellow ]
]
end
I have a NetLogo model. each turtle has two attributes, "closeness" and "deviation_from_oracle". Now let's say there are 1000 agents in the model. The question is, how can I plot the "closeness" against "deviation_from_oracle" ?
It would also be helpful if I can get a csv file from NetLogo that has the value of closeness and deviaiton_from_oracle of all turtles after for example 1500 steps.
I definitely agree with Hugh_Kelley regarding using Behaviorspace to output your values (or custom export functions that might make for easier data cleanup if you're looking to report values for a large dynamic number of turtles- depends on your comfort with your statistical software of choice).
If you do need to plot something on the interface to show your users or something, you may find the plotxy function does what you need. For example, you'll need a plot on the interface called "plot 1" and a single blank pen in that plot called "pen-0".
You can control that plot either by manually setting up its x and y extent or by using the set-plot-... commands as in this setup:
to setup
ca
crt 10
set-current-plot "plot 1"
set-current-plot-pen "pen-0"
set-plot-pen-mode 2
set-plot-x-range 0 17
set-plot-y-range 0 25
reset-ticks
end
If you need to have a value plotted for each of your turtles, you can get the turtles to call plotxy for whatever values you're looking to plot- here I just use their absolute x coordinate and distance to the center as an example:
to go
ask turtles [
rt random 61 - 30
fd 1
set-plot-pen-color color
plotxy ( abs xcor ) distance patch 0 0
]
tick
end
This gives output like:
Where each point was plotted by an individual turtle.
If you want instead some reported mean value, have the observer call plotxy instead- another example that plots the average distance to other turtles and the average distance to center:
to go
plotxy mean-closeness-to-others mean-distance-center
ask turtles [
rt random 61 - 30
fd 1
]
tick
end
to-report mean-closeness-to-others
report mean [ mean map distance sort other turtles ] of turtles
end
to-report mean-distance-center
report mean [ distancexy 0 0 ] of turtles
end
For an output like:
I want to determine the distance the vehicle traveled for comparison with other values, I should use the command / function what to calculate.
for example in the picture, I want to use the function to determine the distance d1, after one-time drive, the distance will be the last .... d2 distance riding is dn
I'm not sure if I understand your question. Perhaps JenB's comment is better for you. But here's a different kind of answer:
A simple way for a turtle to keep track of how far it has traveled is shown in this small example program:
turtles-own [traveled]
to example
clear-all
create-turtles 1
ask turtles [
repeat 5 [
let delta random-float 1.0
fd delta
set traveled traveled + delta
]
print traveled
]
end
Basically, every time the turtle moves, you add the amount it moved to a turtle variable.
This assumes you are using forward to move the turtle. If you are moving the turtle using some other method like setxy or move-to, then you will need different code.
I have a NetLogo model for which I'd like to create different human disturbance scenarios impacting carnivore "prey" across the landscape. Here is my NetLogo landscape of a real protected area. Lighter colored pixels have higher prey values than darker colored pixels.
I'd like to create functions that simulate human disturbance from the edge of the protected area. I'd like to try different functions, such as sigmoidal and exponential decay (i.e., areas closer to edge of protected area have prey pixel values reduced more so than prey pixels farther from the edge).
I can implement some simple functions with the following:
ask patches with [is-park?] ; patches inside the protected area
[
set dist-boundary distance (min-one-of patches with [is-park? = FALSE][distance myself]) ; calculate distance of patch from edge of park
set prey (prey - e ^ (- dist-boundary / 10)) ; scenario 1 human disturbance
set prey (prey - (-0.1 * dist-boundary ^ 2 + 0.9 * dist-boundary)) ; scenario 2 human disturbance
]
However, I'd ideally like to create a set of scenarios such that the total prey reduction across the landscape is equal for each scenario (but distributed across the landscape differently). That would allow me to assess the impact of spatial distribution of human disturbance independent of total magnitude of impact. Any ideas on how to do that would help me a bunch. I've been stuck on this a while now.
It sounds like you are looking for the "diffuse" function.
diffuse "patch-variable" "number"
When you have a human (possibly a turtle) in a patch they can diffuse "negative-influence" a patch variable double I just invented that would remove prey and decrease as it does so.
ask patches with [human?] ;patches with a human create negative influence
[
set negative-influence 8.8; can be any number
]
So you've created negative-influence now we implement the diffusion:
ask patches with [negative-influence>1]
[
set prey prey-(factor*negative-influence); removes some prey
set negative-influence negative-influence*(1-factor); reduce negative-influence
diffuse negative-influence 0.5 ; half gets diffused into neighboring patches
]