Hello Chunyang,

The AIMMS parallel sparse execution engine is designed to handle large amounts of sparse data quickly.  As a consequence, each time the execution of an assignment statement starts, there is some preparation time.

The tactic to write efficient code is to avoid for loops - 
let each execution of an assignment handle as much data as possible.

For the example at hand, Inside the for loops the input identifiers nodeInTime, p_value, and path do not change, and the parameter scenario and set noted_scenario_index are merely used as temporaries.

Instead of re-using these temporaries many times, the following code expands their use
over the domain (n,t) of valid tuples; consequently avoiding repeated assignment preparation times.

I didn't have data, so I cannot predict the time / memory effect of this code change for your application.  Can you let us know?

With kind regards,

Chris.

PS Please find the project I used enclosed. 
PPS Please use AIMMS 25.3 for testing, as it has an important improvement on Indexed Sets that is used in pr_adapt.

Hello Chunyang,

 

The AIMMS parallel sparse execution engine is designed to handle large amounts of sparse data quickly.  As a consequence, each time the execution of an assignment statement starts, there is some preparation time.

The tactic to write efficient code is to avoid for loops - 
let each execution of an assignment handle as much data as possible.

For the example at hand, Inside the for loops the input identifiers nodeInTime, p_value, and path do not change, and the parameter scenario and set noted_scenario_index are merely used as temporaries.

Instead of re-using these temporaries many times, the following code expands their use
over the domain (n,t) of valid tuples; consequently avoiding repeated assignment preparation times.

 

I didn't have data, so I cannot predict the time / memory effect of this code change for your application.  Can you let us know?

With kind regards,

Chris.

PS Please find the project I used enclosed. 
PPS Please use AIMMS 25.3 for testing, as it has an important improvement on Indexed Sets that is used in pr_adapt.

Hi Chris!

Thank you so much for this great reply! I updatted my AIMMS to 25.3 and tried it on my project. Using my previous method, running needed_value(scenario, n, t, i) := value(path(r,i), i); alone would take 49.2 seconds.

Now, with the method you suggested, p_needed_value(i_scen, n, t, i) | ( i_scen in _is_scenarios(n,t) ) := p_value( ep_path( _ep_nodeOfScenario( n, t, i_scen ), i), i); running this line to exert all the indices at the same time would only took 62.17 seconds!!!

It greatly improved the efficiency. With my previous method, I will need 23*1000*49.2 seconds to do what I need. Now I only need 62.17 seconds which is amazing!

Best,

Chunyang