topic Re: getting the values by changing it to reach the target in Mathematical Optimization, Discrete-Event Simulation, and OR

getting the values by changing it to reach the target

Decay2020 — Thu, 03 Feb 2022 15:48:52 GMT

Hi All,

I have below 3 inputs, one is data "Have" and two values "value1" and "Value2". I want to arrive at "final data want" (as described below) such that sum product of col2 and col3 equals a constant by changing value1 and keeping value2 same.

data have ;
infile cards ;
input col1;
cards;
1
2
3
4
5
6
7

;
run;

value1 = -4.56266639

value2 = 0.28179071

I want to calculate col2 as value1 + (value2*col1)

and col3 as EXP(col2)/(1+EXP(col2)) such that sum product of col2 and col3 equals 0.036 by changing the value1.

Final data want:

col1	col2	col3
1	-4.28088	0.013642
2	-3.99908	0.018002
3	-3.71729	0.023723
4	-3.4355	0.031204
5	-3.15371	0.040945
6	-2.87192	0.053559
7	-2.59013	0.069776

Re: getting the values by changing it to reach the target

RobPratt — Fri, 04 Feb 2022 17:16:40 GMT

Here's one way to do it:

%let value1 = -4.56266639;
%let value2 = 0.28179071;

proc optmodel;
   set OBS;
   num col1 {OBS};
   read data have into OBS=[_N_] col1;

   var value1;
   num value2 = &value2;
   impvar col2 {i in OBS} = value1 + value2 * col1[i];
   impvar col3 {i in OBS} = EXP(col2[i]) / (1 + EXP(col2[i]));
   con SumProduct:
      sum {i in OBS} col2[i] * col3[i] = 0.036;

/*   fix value1 = &value1;*/

   solve noobj;
   print value1;
   print col1 col2 col3;
   create data want(drop=i) from [i] col1 col2 col3;
quit;

On my machine, I get:

value1
-1.2786

[1]	col1	col2	col3
1	1	-0.99685	0.26956
2	2	-0.71506	0.32848
3	3	-0.43327	0.39335
4	4	-0.15147	0.46220
5	5	0.13032	0.53253
6	6	0.41211	0.60159
7	7	0.69390	0.66683

To recover the values you showed, you can uncomment the FIX statement to force the solver to return that specific solution:

value1
-4.5627

[1]	col1	col2	col3
1	1	-4.2809	0.013642
2	2	-3.9991	0.018002
3	3	-3.7173	0.023723
4	4	-3.4355	0.031204
5	5	-3.1537	0.040945
6	6	-2.8719	0.053559
7	7	-2.5901	0.069776

But that solution does not satisfy the equality constraint because the resulting SumProduct is -0.789 instead of 0.036.

Re: getting the values by changing it to reach the target

Decay2020 — Fri, 04 Feb 2022 04:56:26 GMT

Thank you for you quick turn around.
I apologize I made an error in the query.
value1 = -4.60875
After optimization it should come as -4.56267, but we start with Value1 as -4.60875

Re: getting the values by changing it to reach the target

RobPratt — Fri, 04 Feb 2022 17:14:17 GMT

It turns out that value1 = -1.2786 is the unique feasible solution. Here is a portion of the plot of SumProduct as a function of value1, with a horizontal reference line at 0.036:

You can initialize the solver at &value1 by changing the VAR statement:

var value1 init &value1;

But initial values to the left of the minimum at about value1 = -2.6 are likely to lead the solver astray.