Don't mine all simulation parameters in each calendar optimization iteration

[david-chapela]

The simulation parameters (json_parameters) don't have to be discovered in each optimization iteration.

I think that currently, at the beginning of each "calendar optimization" iteration, the arrivals, gateway probabilities, etc. are discovered. There is no need to do all of them each time.

JSON Parameters:

• Arrival distribution:
  • Composed of both the arrival_time_distribution and arrival_time_calendar.
  • Only discovered once in the beginning of the main SIMOD optimization (using train log).
  • The same values (distribution and calendar) are going to be used during all the process.
• Gateway probabilities:
  • Composed of only gateway_branching_probabilities.
  • Depends directly on the BPMN model, so they have to be discovered when the model changes (once per iteration of the "structure optimization").
  • In each "structure optimization" iteration, a new model is discovered, so its gateway probabilities are mined (either equiprobable or real) to be able to run Prosimos with it.
  • Once the "structure optimization" ends, there is no need to discover them again, the probabilities of the _bestresult are the same for the rest of SIMOD.
• Resource profiles:
  • Composed of resource_profiles, resource_calendars, and task_resource_distribution.
  • Discovered once before the beginning of the "structure optimization", with default parameter values (this result is used in all the iterations of the "structure optimization", and in the extraneous delays' discovery).
  • Also, discovered once per iteration of the "calendar optimization", with the parameters of the iteration given by fmin, and used in that iteration.
  • Once the "calendar optimization" ends, there is no need to discover them again

Extra: This is for the SIMOD optimization, once all the stages end, to evaluate the model against test log, all these json_parameters are discovered again from train+validation.