Originally posted by: UDOPS

Suggestions on what I can look for in my model to improve its processing speed? According to the profiler tab, cplex only takes 10 seconds to solve the problem, but the "extract" step requires 200 seconds, "self time".
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: UDOPS

More detail: in addition to "extract", the profiler reports these two data structures as taking 77 and 70 seconds to process. Suggestions on making them more efficient?

occheadway SBOccBlockNet = [ <i,t> :
<card({r | <r,i,j,ti,tj,-1> in Paths: t>=ti && t<tj }),
card({r | <r,a,i,ta,ti,-1> in Paths: TrainData[r].headway>0 && t>=ta && t<ti})+
card({r | <r,a,i,ta,ti,-1> in Paths, <r,b,a,tb,ta,-1> in Paths: TrainData[r].headway>1 && t>=tb && t<ta})>
| <i,t> in BlockNet];
occheadway NBOccBlockNet = [ <i,t> :
<card({r | <r,i,j,ti,tj,1> in Paths: t>=ti && t<tj }),
card({r | <r,a,i,ta,ti,1> in Paths: TrainData[r].headway>0 && t>=ta && t<ti})+
card({r | <r,a,i,ta,ti,1> in Paths, <r,b,a,tb,ta,1> in Paths: TrainData[r].headway>1 && t>=tb && t<ta})>
| <i,t> in BlockNet];
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: UDOPS

Follow up:

I have to say I have spent a chunk of time using the profiler and I am very pleased with the insight it provides. I have been re-organizing the creation and re-use of enumerated sets, and have been able to cut my processing time significantly.

Basically, I looked for places where the model steps through the same set repeatedly, and saved the results to an interim set for re-use later. Generally, I cut my model extraction time in half.
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: SystemAdmin

If TrainData contains few values >0, you can perhaps try something like the following formulation.
The idea is to reduce the domain of the iterations.
(Please check my formulation is still equivalent to yours.)
{T3} smallPaths1 = {<r,i,j,ti,tj> | <r,i,j,ti,tj,-1> in Paths: tj>ti};
{T3} small00 = {<r,i,j,ti,tj> | <r,i,j,ti,tj> in smallPaths1: TrainData[r].headway>0};
{T3} small01 = {<r,i,j,ti,tj> | <r,i,j,ti,tj> in small00: TrainData[r].headway>1};
{T3} small02 = {<r,i,j,ti,tj> | <r,i,j,ti,tj,-1> in Paths: TrainData[r].headway>1};
{T3} smallPaths2 = {<r,i,j,ti,tj> | <r,i,j,ti,tj,1> in Paths: tj>ti};
{T3} small00 = {<r,i,j,ti,tj> | <r,i,j,ti,tj> in smallPaths2: TrainData[r].headway>0};
{T3} small11 = {<r,i,j,ti,tj> | <r,i,j,ti,tj> in small00: TrainData[r].headway>1};
{T3} small22 = {<r,i,j,ti,tj> | <r,i,j,ti,tj,1> in Paths: TrainData[r].headway>1};

T1 A1BlockNet = [ <i,t> :
<card({r | <r,i,j,ti,tj> in smallPaths1: tj>t>=ti}),
card({r | <r,a,i,ta,ti> in small00: ti>t>=ta})+
card({r | <r,b,a,tb,ta> in small01, <r,a,i,ta,ti> in small02: ta>t>=tb})>
| <i,t> in BlockNet];

T1 A2BlockNet = [ <i,t> :
<card({r | <r,i,j,ti,tj> in smallPaths2: tj>t>=ti}),
card({r | <r,a,i,ta,ti> in small00: ti>t>=ta})+
card({r | <r,b,a,tb,ta> in small11, <r,a,i,ta,ti> in small22: ta>t>=tb})>
| <i,t> in BlockNet];
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: UDOPS

Thanks!

Currently my data is all or nothing, but I envision future problems where the parameter headway varies quite a lot.
#DecisionOptimization
#OPLusingCPLEXOptimizer