How can I collect the “ids” of directed links using the NetLogo Behavior Space?

问题

We have stored a lot of data as attributes of links in our NetLogo model. When I use the Behavior Space to design experiments and to direct data collection, I specify the "[attribute] of link" to be extracted. However, in the csv file I cannot see the id of the link, which is a essential to understand the data. How can I collect the ids of links in my results? From what I understand there is no primitive to enable this command.

回答1:

Whenever you want to extract information from individual agents (including links) using BehaviorSpace, a nice way to do it is to use the csv extension as explained in this answer:

https://stackoverflow.com/a/52406247/487946

The general idea is that we can embed csv into our csv, and then use a read_csv-like function in R (or Python or Julia or whatever) to extract the "inner csv" from our BehaviorSpace results.

In the case of links, it would be useful to include the who number of each end of the link to uniquely identify it. (This is one of the very few cases where I'll advocate using a who number for anything.)

Let's take this silly example model:

extensions [ csv ]
links-own [ attribute ]

to setup
  clear-all
  create-turtles 3 [
    create-links-with other turtles [
      set attribute random-float 1
    ]
  ]
  reset-ticks
end

to go
  ask links [ set attribute attribute * 0.5 ]
  tick
end

It just creates three turtles with links between them, sets the attribute of the link to a random number and repeatedly halves that number as the model ticks.

To generate the csv that we will embed into our BehaviorSpace results, we write the following reporter:

to-report link-attributes-csv
  report csv:to-string
    fput ["who1" "who2" "attribute" ] 
    [ (list [ who ] of end1 [ who ] of end2 attribute) ] of links
end

If you try it out in the command center after running setup, it will output something like this:

observer> setup
observer> print link-attributes-csv
who1,who2,attribute
0,1,0.9409784968740699
1,2,0.9079884204004846
0,2,0.9070292656950991

As you can see, we have a neat little csv table, where each row represents a particular link, identified by the who number of the turtles it connects.

Since this reporter reports a string (and it's ok for this string to contain line breaks), we can use it directly in a BehaviorSpace experiment:

Running this experiment (with "table output") gives the following output file:

"BehaviorSpace results (NetLogo 6.1.1)"
"link-attributes-example.nlogo"
"experiment"
"10/16/2019 11:00:12:495 +0100"
"min-pxcor","max-pxcor","min-pycor","max-pycor"
"-16","16","-16","16"
"[run number]","[step]","link-attributes"
"1","0","who1,who2,attribute
1,2,0.15670083797389645
0,2,0.40055350697928993
0,1,0.34892645306446335"
"2","0","who1,who2,attribute
0,1,0.2831244347856665
1,2,0.27721328746715357
0,2,0.5221352362751627"
"2","1","who1,who2,attribute
0,1,0.14156221739283326
0,2,0.26106761813758134
1,2,0.13860664373357678"
"1","1","who1,who2,attribute
0,2,0.20027675348964497
1,2,0.07835041898694822
0,1,0.17446322653223167"
"1","2","who1,who2,attribute
1,2,0.03917520949347411
0,2,0.10013837674482248
0,1,0.08723161326611584"
"2","2","who1,who2,attribute
1,2,0.06930332186678839
0,1,0.07078110869641663
0,2,0.13053380906879067"

It looks at bit weird with all the line breaks, but your data analysis tools should be able to deal with that. Here is how to handle this using R and the Tidyverse:

library(tidyverse)
df <-
  read_csv("experiment-table.csv", skip = 6) %>%
  mutate(`link-attributes` = map(`link-attributes`, read_csv)) %>%
  unnest()

The purrr::map and tidyr::unnest functions are the key ones. I won't explain them here, but its worth looking them up and familiarizing yourself with them.

Our final result looks like this:

# A tibble: 18 x 5
   `[run number]` `[step]`  who1  who2 attribute
            <dbl>    <dbl> <dbl> <dbl>     <dbl>
 1              1        0     1     2    0.157 
 2              1        0     0     2    0.401 
 3              1        0     0     1    0.349 
 4              2        0     0     1    0.283 
 5              2        0     1     2    0.277 
 6              2        0     0     2    0.522 
 7              2        1     0     1    0.142 
 8              2        1     0     2    0.261 
 9              2        1     1     2    0.139 
10              1        1     0     2    0.200 
11              1        1     1     2    0.0784
12              1        1     0     1    0.174 
13              1        2     1     2    0.0392
14              1        2     0     2    0.100 
15              1        2     0     1    0.0872
16              2        2     1     2    0.0693
17              2        2     0     1    0.0708
18              2        2     0     2    0.131

I hope this helps.

回答2:

Like patches, links are identified by two numbers, which are the who numbers of the two ends. You can either save the string representation of the link (e.g., (link 0 1)) or extract the numbers as a list (or separately). E.g.,

to test
  ca
  crt 2
  ask turtle 0 [create-link-with turtle 1]
  print link 0 1
  ask link 0 1 [let id sort [who] of both-ends print id] ; a list
  ask link 0 1 [let id sort [who] of both-ends
    print (word item 0 id "-" item 1 id)
  ] ; a string
end

来源：https://stackoverflow.com/questions/58398560/how-can-i-collect-the-ids-of-directed-links-using-the-netlogo-behavior-space

标签

netlogo

data-collection

behaviorspace