topic Re: random access to variables in observations in Mathematical Optimization, Discrete-Event Simulation, and OR

random access to variables in observations

jdemaio — Thu, 01 Feb 2018 15:19:41 GMT

Hello,

I've playing around with Proc OptGraph recently. I have a data set consisting of district representatives in CA and the committees that they serve on. Here are the first 10 observations sorted by committee assignment.

Obs   District   Name   Committee Assignment
1   16th   Costa, Jim   Agriculture
2   10th   Denham, Jeff   Agriculture
3   1st   LaMalfa, Doug   Agriculture
4   20th   Panetta, Jimmy   Agriculture
5   31st   Aguilar, Pete   Appropriations
6   42nd   Calvert, Ken   Appropriations
7   13th   Lee, Barbara   Appropriations
8   40th   Roybal-Allard, Lucille   Appropriations
9   21st   Valadao, David   Appropriations
10   24th   Carbajal, Salud   Armed Services

I've analyzed the bipartite graph whose nodes come from the name and committee assignment variables. Edges appear from name to committee assignment if that member serves on said committee.

I now want to create a new graph whose nodes are the names and an edge exists between two names if they sit on a common committee. For example, Agriculture will generate 4 choose 2= 6 edges in the new graph, one of which is Costa-Denham. So, I want to create a new data set from my current data set with two variables:rep1 and rep2. I think that I want to use nested for loops to check committee assignment for every pair of observations and if they match then output the observation names into my new data set as rep1 and rep2. My first 7 observations in the new data set would be:

Obs rep1 rep2

1 Costa Denham

2 Costa LaMalfa

3 Costa Panetta

4 Denham LaMalfa

5 Denham Panetta

6. LaMalfa Panetta

7 Aguilar Calvert

What I don't know how to do, if possible in SAS, is look at the specific entries in observations to compare and then write to a new data set. Suggestions?

Thanks,

Joe

Re: random access to variables in observations

RobPratt — Thu, 01 Feb 2018 17:30:24 GMT

proc optmodel;
   set <str,str> ARCS;
   read data bipartite into ARCS=[Name CommitteeAssignment];
   create data outdata from [rep1 rep2]=(setof {<n1,c> in ARCS, <n2,(c)> in ARCS: n1 < n2} <n1,n2>);
quit;

Re: random access to variables in observations

jdemaio — Thu, 01 Feb 2018 18:45:09 GMT

Works perfectly.

Thanks!

Joe

Re: random access to variables in observations

RobPratt — Thu, 01 Feb 2018 18:56:20 GMT

And here's another approach that uses PROC OPTGRAPH:

data names(keep=node source);
   set bipartite(rename=(name=node));
   source = 1;
run;
proc optgraph links=bipartite data_nodes_sub=names;
   links_var from=Name to=CommitteeAssignment;
   shortpath out_weights=outdata(rename=(source=rep1 sink=rep2) where=(rep1 < rep2 and path_weight=2));
run;

Re: random access to variables in observations

RobPratt — Fri, 02 Feb 2018 15:16:41 GMT

And another way with PROC SQL:

proc sql;
   create table outdata as
   select a.Name as rep1, b.Name as rep2
   from bipartite as a full outer join bipartite as b
   on a.CommitteeAssignment = b.CommitteeAssignment
   where rep1 < rep2;
quit;