How to create a graph object in NetLogo - netlogo

I'd like to use the NetLogo's R extension to send a graph object into R, then use the iGraph package to calculate and return some metrics. iGraph can create a graph from an adjacency matrix or an edgelist (there are other options too). The graph I'd like to send to R is just the links among an agentset. Does anyone know how to do this? The NW:save-matrix will export an adjacency matrix to a file. Do I need to do this and then read the file back into R or is there a more direct way?
As always, thank you!

What I have done in the past is construct the network within NetLogo, export the network to R, calculate the network metrics in R, and then retrieve the metrics. The relevant code I used in one of those projects is:
to export-nw2r
; create file with useful graph format
nw:set-context people links
let filename (word "Networks/netlogo" behaviorspace-run-number ".gml")
export-simple-gml filename
;; reset the R-workspace
r:clearLocal
let dir pathdir:get-model
r:eval "library(igraph)"
; read network in R
set filename (word dir "/" filename)
r:put "fn" filename
r:eval "gg <- read_graph(file = fn, format = 'gml')"
r:eval "V(gg)$name <- V(gg)$id" ; gml uses 'id', but igraph uses 'name'
r:eval "if (file.exists(fn)) file.remove(fn)"
end
to calc-network-properties
r:eval "library(ineq)"
; network size
set sizeN count people
set sizeE count links
output-type "Nodes: " output-print sizeN
output-type "Edges: " output-print sizeE
; calculate degree properties
r:eval "degs <- degree(gg)"
r:eval "aveDeg <- mean(degs)"
set aveDeg r:get "aveDeg"
output-type "Mean Degree: " output-print precision aveDeg 2
r:eval "giniDeg <- ineq(degs, type = \"Gini\")"
set giniDeg r:get "giniDeg"
output-type "Gini of Degree: " output-print precision giniDeg 2
; calculate transitivity properties
r:eval "lccs <- transitivity(gg, type = \"localundirected\")"
r:eval "aveCC <- mean(lccs, na.rm = TRUE)"
set aveCC r:get "aveCC"
output-type "Mean Clustering: " output-print precision aveCC 2
r:eval "trans <- transitivity(gg, type = \"undirected\")"
set trans r:get "trans"
output-type "Transitivity: " output-print precision trans 2
; paths and betweenness
r:eval "paths <- distances(gg)"
r:eval "paths <- paths[upper.tri(paths)]"
r:eval "avePath <- mean(paths)"
set avePath r:get "avePath"
output-type "Mean Shortest Path: " output-print precision avePath 2
r:eval "diam <- max(paths)"
set diam r:get "diam"
output-type "Max Shortest Path: " output-print diam
r:eval "giniPaths <- ineq(paths, type = \"Gini\")"
set giniPaths r:get "giniPaths"
output-type "Gini of Paths: " output-print precision giniPaths 2
r:eval "btws <- betweenness(gg)"
r:eval "giniBtwn <- ineq(btws, type = \"Gini\")"
set giniBtwn r:get "giniBtwn"
output-type "Gini of Betweenness (V): " output-print precision giniBtwn 2
end

Related

NetLogo database resultlist loading problem

I am trying to get some integer values from database (almost 30.000 rows) and it gives error "error while observer running READ-FROM-STRING in procedure GET-IMAGEDB Expected a constant." I considered it may happen because of requiring waiting time to not set a null value to variable of NetLogo and added "print" lines to delay the process. It worked for small sized datasets for instance 30.000 rows. However, when the dataset size increased to 250.000 rows, now I am getting same error again. I am not sure about reason of the error. I have also tried to get all rows from the dataset with single query and set it to NetLogo list, but still gives same error and extremely slow. I have attached the code. Thank you for any help.
let total ( (width + 1) * (height + 1) )
let counter 0
while [ counter < total ]
[ set query "select x, y, r, g, b, background from colorinfo where pixelId = \""
set query (word query counter "\"" )
let resultList (mysql:executeQuery db query)
let stringList item 0 resultList
print (word "wait 1")
let x read-from-string item 0 stringList
print (word "wait 2")
let y read-from-string item 1 stringList
let r read-from-string item 2 stringList
let g read-from-string item 3 stringList
let b read-from-string item 4 stringList
let background read-from-string item 5 stringList
ask patch (x + width + 2) ( y) [ set pcolor (list r g b) ]
print counter
set counter (counter + 1)
set stringList [ ]
set resultList [ ]
]

Multi-scale landscape in Netlogo (small patches and larger patch groupings)

I am trying to represent a multi-scale environment where I have large patches that represent high-value areas in the landscape and smaller patches that have local information. E.g. I want to have snow data at a 1km^2 scale but I also want to have larger patches (9km^2) that summarize large-scale information. Each of my large patches has a variable value that is different from its neighbors but the variable value may be repeated throughout the landscape in other patches. I am looking for the most straightforward way for my turtles to identify the difference between the large-scale patches. I had thought of creating patch-sets but I am not sure how to get around the issue of variable values repeating in different patches. Any help is much appreciated.
EDIT: I have created a raster with equal patch structure as the large-scale raster and assigned "patch-id's" using this, so that there is no longer variable repetition in the world. I am still struggling with getting turtles to identify these larger patches as grouped entities.
You commented on my first answer
My main issue is that I need to run a "find max-one-of
neigboring-large-patches [large-scale-variable]" so I need my turtles
to understand what the neighboring large-patches are and be able to
read them as units, if that makes sense. I can't quite figure out how
to incorporate that into your answer, any thoughts?
Here's how to do that. This code is fast and sloppy but it illustrates the point.
Let the large-regions have x and y values, generated during creation. Basically, these store the column and row numbers of the grid of large regions that covers the viewport.
breed [ large-regions large-region ]
large-regions-own [
terrain
region-color
population
x
y
]
Then, conceptually, the neighbors of a region will have x and y values within +/- 1 of the region's x and y values, so you can identify them that way.
To simplify coding at the expense of space, when I generated the regions I also stored the unique identifier (who) of that region and its x and y values into every patch in that region, in variables lrx and lry.
patches-own [
large-region-who
lrx
lry
]
The heart of finding the neighboring large-region with the max value of population as you requested follows. I coded this for speed in debugging , not for elegance, so it can be greatly cleaned up. The full source code has many print statements that effectively comment each step in solving your requested search.
This looks around (patch 0 0), finds the info on the large region's x and y from that patch, generates an agent-set of large-regions with nearby x and y values, does a max [population] search on that set to extract the region with the highest population. It also colors the asking patch black, the local large-region blue, and the maximum population neighbor red.
It mostly works -- the large regions are offset by one patch from where they should be -- but this illustrates the point. Run setup and go and see for yourself.
Here's the (ugly) code to play with. Interesting problem. You can easily extend this to small regions as well, and have both working at the same time. Enjoy!
globals [
large-region-size
]
breed [ large-regions large-region ]
large-regions-own [
terrain
region-color
population
x
y
]
patches-own [
large-region-who
lrx
lry
]
to setup
clear-all
set large-region-size 5
no-display
make-large-regions
ask patches [ set pcolor white ]
display
ask large-regions [ set hidden? true]
print (word " hilly region count: " count large-regions with [terrain = "hilly"] )
;; print (word " deep snow count: " count small-regions with [snow-cover > 75])
reset-ticks
end
to go
ask patches [ set pcolor white]
; ;; lets examine the large-regions
; print " large region xvals "
; let xvals [ ]
; ask large-regions [ set xvals fput x xvals ]
; set xvals remove-duplicates xvals
; show xvals
; print " "
; print " patch lrx values: "
; set xvals [ ]
; ask patches [ set xvals fput lrx xvals ]
; set xvals remove-duplicates xvals
; show xvals
; print "========================================="
print " let's examine large-regions around the patch at 0 0 "
let x-spot 0
let y-spot 0
print ( word " looking for large-regions with max population bordering the following patch " x-spot " " y-spot)
; ask n-of 1 patches [ set x-spot pxcor set y-spot pycor print (word "selected patch " x-spot ", " y-spot )]
let home-who [ large-region-who] of patch x-spot y-spot
print (word "home-region-who is " home-who)
print " "
;; thinking ahead, we have coded the x and y values of the large region around us directly into the patch variables
let home-x [ lrx ] of patch x-spot y-spot
let home-y [ lry ] of patch x-spot y-spot
print (word "this blue home region has x=" home-x " and y=" home-y )
ask patches with [lrx = home-x and lry = home-y] [ set pcolor blue ]
ask patch x-spot y-spot [ set pcolor black ]
let home-neighbor-set large-regions with [
( x >= ( home-x - 1 )) and ( x <= ( home-x + 1) ) and (y >= ( home-y - 1 ) ) and ( y <= ( home-y + 1 ) ) ]
print "count of home-neighbor-set is "
print count large-regions with [
( x >= ( home-x - 1 )) and ( x <= ( home-x + 1) ) and (y >= ( home-y - 1 ) ) and ( y <= ( home-y + 1) ) ]
print " "
print "here is that set "
show home-neighbor-set
print " "
ask home-neighbor-set [ print (word "Large region with who = " who " has population " population )]
let big-boy max-one-of home-neighbor-set [ population]
show big-boy
print ( word " Neighboring red large-region with largest population is " big-boy " with population " [population] of big-boy )
let bbx 0
let bby 0
let bwho 0
ask big-boy [ set bbx x set bby y set bwho who]
ask patches with [lrx = bbx and lry = bby] [ set pcolor red ]
tick
end
to make-large-regions ;; for testing
let px min-pxcor
let py min-pycor
let region-id -1 ;; missing
let mysize large-region-size
let stopper 0
while [px < max-pxcor] [
while [py < max-pycor] [
if stopper > 300 [ stop ] ;; stops making large regions
set stopper stopper + 1
let xcode round ( ( px + 1) / 5)
let ycode round ( ( py + 1) / 5)
;; make a new region
let decolor one-of [ red blue yellow green ]
create-large-regions 1 [
set terrain one-of ["hilly" "flat" "mountain" "water" "swamp"]
set region-id who
set population random 1000
set x xcode
set y ycode
set region-color decolor
]
;; large region is defined, update the patches in that region
ask patches with [ (abs (pxcor - px) < (mysize / 2) )
and (abs (pycor - py) < (mysize / 2) )] [
set pcolor decolor
set large-region-who region-id
set lrx xcode
set lry ycode
]
set py py + mysize
]
if py > max-pycor [
set py min-pycor
set px px + mysize]
]
end
This may not be the best way, but I think it would work. You could let regions own several variables, such as "large-region-unique-id" and "small-region-unique-id" and make one pass where you set all these variables. Then a turtle would only have to look at a patch to know what small and large region it is part of.
If you also made a breed of agents called "regions" (say), you could have regions-own variables and have a unique-region-id. ( actually, the agent's who number would work
for that)
That should encode the information so that a moving turtle could easily look up relevant information.
breed [ large-regions large-region ]
large-regions-own [
terrain-type
large-scale-variables
...
(who)
]
breed [ small-regions small-region ]
small-regions-own [
snow-cover
small-scale-variables
...
(who)
]
patches-own [
large-scale-region-who ;; the id (who) of the large-scale-region the patch is in
small-scale-region-who ;; the id (who) of the small-scale-region the patch is in
...
]
Then a turtle could ask a patch for the relevant who information and use it to look up data from the larger "patches".
Here's what that might look like
print (word " hilly region count: " count large-regions with [terrain = "hilly"] )
print (word " deep snow count: " count small-regions with [snow-cover > 75])
;; how about highlighting patches that are mountainous with deep snow?
no-display
ask patches [
set terrain-type ""
set my-snow-cover -1
set srw small-scale-region-who
if srw > 0 [set my-snow-cover [snow-cover] of (small-region srw)]
set lrw large-scale-region-who
if lrw > 0
[ set terrain-type [terrain] of large-region lrw]
if-else (terrain-type = "mountain") and (my-snow-cover > 75)
[ set pcolor white ]
[ set pcolor black ]
]
display
print " The mountainous terrain with deep snow-cover is shown in white "

Changing range of variables netlogo

I have a variable:
ask group [set means-one groupmeans + resources-agent ]
I want to ask netlogo to constrain the variable between 1 to 99.
How?
Just to simplify Alan's first answer to remove the if statements:
You could do:
ask group [set means-one (max (list 1 (min (list 99 groupmeans + resources-agent))) ]
Your question is not entirely clear. What do you mean by "change the range"? If you mean to clip extreme values, you can do it like this:
to-report clip [#x #min #max]
if (#x < #min) [report #min]
if (#x > #max) [report #max]
report #x
end
Then you can ask turtles [set means-one clip means-one 1 99]. Otoh, if you actually want to rescale all existing values linearly into your new range, you could do the following:
to rescale-all-means-one
let _newmin 1
let _newmax 99
let _newrange (_newmax - _newmin)
let _lst [means-one] of turtles
let _min min _lst
let _max max _lst
let _range (_max - _min)
ask turtles [
let _scale (means-one - _min) / _range
let _scaled (_newmin + _scale * _newrange)
set means-one _scaled
]
end

How make a list of cumulative sum in netlogo

How can i make a list of cumulative sum of a other list?
i tried it that way:
;;all temperatrue-values around the turtle saved in list
set temperature_values (list [(output-heat + 1)^ Freedom] of neighbors)
;;build cumulative value of temperatures and put each value in list
let tempsum 0
set tempsum_list []
foreach temperature_values
[set tempsum (tempsum + ? )
set tempsum_list fput tempsum tempsum_list
]
but it doesn't work. can anyone fix this problem? it says that "+ excepted a input but gets a list instead".
your code for a cumulative sum works (except that I think you need lput rather than fput. You can see it with this:
to test
let ll [1 2 3 4]
let tempsum 0
let tempsum_list []
foreach ll
[ set tempsum (tempsum + ? )
set tempsum_list lput tempsum tempsum_list
]
print tempsum_list
end
Did the error highlight the line set temperature_values (list [(output-heat + 1)^ Freedom] of neighbors)? Try putting a space after between ) and ^. NetLogo is picky about space around mathematical operators.
As Jen suggested, you can use foreach. Another nice approach is reduce:
to-report partial-sums [#lst]
set #lst (fput [0] #lst) ;;prepare for reduce
report butfirst reduce [lput (?2 + last ?1) ?1] #lst
end
Similar to Alan's solution (Just an update for the recent version of NetLogo that replaces ? with -> for anonymous procedures.)
to-report partial-sums [lst]
report butfirst reduce [[result-so-far next-item] -> lput (next-item + last
result-so-far) result-so-far] fput [0] lst
end
This is like Alan's solution, just abstracted a bit further. (Perhaps too far, depending on your taste! I like JenB's solution as well.)
Let's first define a thing like reduce, but that keeps all the intermediate results:
to-report scan [fn xs]
report reduce [lput (runresult fn ?2 last ?1) ?1]
(fput (list first xs) butfirst xs)
end
Now we can use it to compute partial sums:
observer> show scan task + [1 2 3 4 5]
observer: [1 3 6 10 15]
but we are also free to swap in a different operation:
observer> show scan task * [1 2 3 4 5]
observer: [1 2 6 24 120]

Nested foreach in NetLogo

I am trying to calculate the Gini coefficient of a set of numbers. The Gini coefficient is half the mean absolute difference. That is, for every possible pair of numbers in the list, I need to take their absolute difference and add these differences together (and some other stuff). This is my code
to-report calc-Gini [list-Values]
let sumdiff 0
foreach list-Values
[ foreach list-Values
[ set sumdiff sumdiff + abs ( ?1 - ?2 )
]
]
report 0.5 * sumdiff / (mean list-Values * (length list-Values) ^ 2)
end
When I test it (eg show calc-Gini (list 1 2 3)) I get an error "task expected 2 inputs, but only got 1" on the second foreach.
I think the problem is that NetLogo wants to run through the foreach loops simultaneously. So if the list length is N, then it creates only N pairs (that is, first item in list1 and first item in list2, then the second item in each list etc) which is where the requirement for equal length lists comes from. But I need it to work with the N^2 pairs obtained by crossing the lists.
How can I make the nested foreach do what I want and/or is some other primitive more appropriate?
NetLogo doesn't have a mechanism for binding ?1 and ?2 to an outer and an inner task. When it sees ?1 and ?2 in your code, it expects that both inputs will come from the inner task. And since the inner foreach only provides one input, NetLogo complains.
You can get around that problem by simply assigning the input of the outer foreach to a local variable:
to-report calc-Gini [list-Values]
let sumdiff 0
foreach list-Values
[ let v ?
foreach list-Values
[ set sumdiff sumdiff + abs ( v - ? )
]
]
report 0.5 * sumdiff / (mean list-Values * (length list-Values) ^ 2)
end
That being said, here is an alternative implementation:
to-report calc-gini [ xs ]
report 0.5 * sum map [ sum-diff ? xs ] xs / (mean xs * (length xs) ^ 2)
end
to-report sum-diff [ x xs ]
report sum map [ abs (x - ?) ] xs
end
I can't solve your nested foreach approach, but this might be an alternative way to do your calculation:
If you use ordered data, you can use this equation for the Gini coefficient (given a vector $y$ with $y_i$, $i=1,...,n$)
$$ G(y) = \frac{1}{n} (n + 1 - 2 * \frac{ \sum_{i=1}^{n} (n + 1 - i) y_{i} }{ \sum_{i=1}^{n} y_i} $$
and the following reporter should deliver the result in NetLogo:
to-report calc-Gini [list-Values]
let values sort list-Values ; making sure values are in a non-decreasing order
let n length values
let i 1
let numerator []
foreach values
[ set numerator lput ( (n + 1 - i) * ? ) numerator
set i i + 1
]
report 1 / n * ( n + 1 - 2 * (sum(numerator) / sum(values)) )
end