Optimize Tab

Values that can be set and their meanings are as follows.

Sampler

• Sets the algorithm to perform the optimization. The following types are available.
• All are provided by Optuna.
  1. Bayesian optimization(TPE)
  2. Bayesian optimization(GP)
  3. Genetic algorithm(NSGA-II)
  4. Evolution strategy(CMA-ES)
  5. Quasi-MonteCarlo
  6. Random
  7. Grid

Number of trial

• This number of trials will be performed.
• If the grid sampler is selected, the calculation is performed by dividing each entered Variable by this number.
  • Note that the number of calculations is (Number of trial) to the power of (Number of Variable).

Timeout(sec)

• After the time set here elapses, optimization stops.
• If 0 is input, no stop by time is performed.

Study Name Group (new in v0.6)

• When starting a new optimization, enter the name of the optimization in "Create New Study" and uncheck the other checkboxes.
• To continue an existing optimization, check the "Continue" checkbox and select the study you wish to continue from "Existing Study".
• You can also copy an existing optimization result and continue resuming it under a different StudyName.
  • Check the "Continue" and "Copy" checkboxes, select the study you wish to copy from "Existing Study" and enter the name of the copied study in the "Create New Study" field.
  • This is useful, for example, if you want to run TPE and GP, with the same initial sampling results, respectively.

Run & Stop

• RunOptimize
  • Push the button to perform the optimization.
  • If the StudyName you set already exists in the results file, the optimization is performed as a continuation of that result.
• Stop
  • Force optimization to stop.
  • Even when stopped, the system automatically saves the results up to the most recent evaluation.

Realtime Result (new in v0.6)

• The number of the currently running trial and the optimization status are displayed at the bottom of the progress bar in real time as the optimization runs.
• The state indication of optimization depends on the number of objective functions.
  • For single objective optimization, the minimum value is displayed.
  • In the case of multi-objective optimization, the ratio from one previous step of Hypervolume calculated from the results of the first two objective functions is displayed.
    • If the Hypervolume Ratio frequently exceeds 1, it is not yet converged.