Stopping criteria for default method about calibrar HOT 3 CLOSED

For the AHR-ES method, it means the "step-size reduction" has reached below 0.001 than the original one. Currently, there's no optimal stopping criteria, the 0.001 is totally arbitrary and may be to harsh for some models and not enough for others. A recommended approach can be to manually check the convergency and restart with a lower control$convergence if necessary.

from calibrar.

from calibrar.

I just released in CRAN the version 0.9 of the package. This includes a more standard automatic stopping criteria for the AHR-ES method:

• 0: maxit/maxgen only
• 1: 1 OR max step reduction
• 2: relative tolerance on value (smoothing for AHR-ES)
• 3: maximum number of generations without improvement of reltol.

The previous method was number 1, and the new default is method 2. It will smooth the history of the objective value (due to the inherent randomness of the method) and after that will apply the same criteria used in derivative based methods (if the difference between iterations/generations is below the relative tolerance, the algorithm stops).

You can check the .smooth_stop2() function for more details, but the key part is:

  x_tol = reltol * (abs(x0) + reltol) # x0 is the current value of the objective function
  test = abs(dx) < x_tol # dx is the difference in value of the objective function between generations

from calibrar.