Specify the missing value strategy and stopping rules for the Response Optimizer

Response Optimizer > Optimizer

These options are available for Random Forests^® models, MARS^® models, and TreeNet^® models.

Missing Value Strategy

If the specifications in Options include the consideration of missing values, then the missing value strategy affects how the search works. Usually, Dynamic works well.

Dynamic: The dynamic strategy uses an adaptive strategy to find an optimal solution with or without missing values. The dynamic strategy models the probability of missing for individual predictor variables that had missing values in the training data. The probability of missing depends on the desirability of candidate solutions in the search. For example, suppose the optimal solution occurs when X1 is missing, X2 = -1.5, X3 is missing, and X4 is one of {“red”, “blue”, “green”}. Then, as the optimizer searches, the algorithm tends to increase the probability that X1 and X3 are set to be missing when X2 approaches -1.5 and X4 is one of {“red”, “blue”, “green”}.
Heuristic: The heuristic strategy uses a prefabricated procedure to find an optimal solution with or without missing values. Consider the heuristic procedure when the dynamic strategy consumes too many computing resources.

Stopping Rules

Specify when to stop the search for an optimal solution. Ideally, the search finds a solution with a desirability of 1 and the values of the predictors are satisfactory. Usually, you lengthen the search to try to find a solution with a higher desirability.

Time in minutes exceeds: Increase the time to try more solutions. Enter a value of 0 or greater.
Small values let you get a solution quickly, such as if you want to show sample output but do not need a solution with high desirability. For example, a value of 0 provides a solution from the first iteration.
Note
In the web app, 29 is the maximum value.
Iterations exceed: Usually, you set a time instead of a number of iterations because the time to complete a number of iterations varies from data set to data set. Specify a greater number of iterations to try more solutions.
Small values let you get a solution quickly, such as if you want to show sample output but do not need a solution with high desirability. For example, a value of 0 provides a solution from the first iteration.

Composite desirability is greater than or equal to: Ideally, the search finds a solution with a desirability of 1 and the values of the predictors are satisfactory.; Select this option and decrease the value from 1 to try to shorten the search. The search completes at the first iteration where at least 1 solution has the minimum desirability.; Deselect this option to extend the search until the search reaches the limit for the time or for the number of iterations. The search continues even if many solutions reach the minimum desirability. Usually, you deselect this option to see solutions with more variation in the values of the predictors.

Specify the missing value strategy and stopping rules for the Response Optimizer

Missing Value Strategy

Stopping Rules

Note