The values of this parameter are
- PredictorPermutation -- Uniform permutation of the predictor variable. Choose this option if you believe the target variable follows the tree and the predictor variable does not. This option is typically the most relevant and we have found it to
be the most robust.
- TargetPermutation -- Uniform permutation of the target variable. This option is included for completeness, but should generally not be used. The underlying assumption is that the target variable doesn't follow the tree, which violates the assumption
- PredictorParametric -- Parametric bootstrap of the predictor variable, using the tree and the parameters learned from the single variable model. Choose this option if you believe the predictor variable clostely follows the tree. If it does not, choosing
this option will lead to conservative q-value estimates.
- TargetParametric -- Parametric bootstrap of the target variable, using the tree and the parameters learned from the single variable model. This option may be appropriate if the target data tightly follow the tree and nothing is known about the structure
of the predictor variable.