meta data for this page
  •  
Action disabled: register

Feature request


Comments and feedback will help us focus on the most useful features in demand. What new features you would like to see in jEPlus? For example, how many of you would want to see Sensitivity Analysis tools built into jEPlus? Or, is integrated charting function more useful to you?

[This discussion board has been closed. Please post new topics on the jeplus.org forums]

Discussion

Daniyal Dely, 2014/08/16 17:48

It is So Much Nice to make an Easy link Between jEplus and Other ImPortant Tools Like MATLAB,….Such as MLE+.

Liam Hendricken, 2014/09/03 16:53

Dr. Zhang and Dr. Korolija,

I just wanted to clarify the process when using jEPlus+EA - I understand that pre-calculated results can be used; does jEPlus+EA reference completed simulations during optimization, and skip a simulation if results already exist for a particular case? I notice for large optimization problems (many generations), later generations complete much more quickly, and I assume it is because the EA algorithm uses existing simulation results. Is this assumption correct?

After completing one optimization problem, if I reference the same output folder for a new optimization problem (i.e., new objective functions or crossover rate), does jEPlus+EA do the same thing? Will it look for completed simulations and use existing results in each job folder?

Thanks,

Liam

Yi Zhang, 2014/09/08 10:53

Dear Liam,

If you are referring to the “use pre-calculated results” for simulation options, it is mainly for demonstrating and testing optimisation algorithms. You can take a look at the example named “ET-PreCalc” in the package and see how it works.

Caching functions indeed exist to skip a simulation if results already exist. This is the reason why you saw generations simulating faster with the progression of optimisation. The implementations are different when simulations are executed locally or on a JESS server. For local execution, jE+EA maintains a catalog of evaluated cases to avoid unnecessary simulation calls. The catalog is in memory and is reset each time you start optimisation. So, running the same project in the same folder does not reuse results. (The reason for this is that jE+EA is not smart enough to tell if an IDF model has changed or not between projects.)

When simulations are run remotely on a JESS server, caching is done on the server side. This is disabled by default, but can be enabled by setting a flag (not yet exposed on the GUI) in the project.

Do you feel it is useful to reuse simulation results in existing job folders between optimisation runs? Apparently it will save simulation time, but make it error-prone at the same time.

Yi