By default, the number of features to use in each split is set to be the square root of the total number of
features. This can be changed through the features selection option. If features
selection is set to custom, then the value of maximum features will be used.
If features selection is set to proportion, then the value of proportion features is used.
Otherwise, the value of the maximum features and proportion features options are ignored.
By default, bootstrap sampling is used, with the number of bootstrap samples set through the bootstrap samples
factor option. However, if the value of the bootstrap option is set to no then no bootstrapping is
done, i.e., each tree uses the full dataset.
Additionally, histograms are turned on by default in decision forest classifiers as they typically speed up training significantly.
The maximum number of bins used for each feature is controlled by the option maximum bins. Its default value of 256 is enough to not
lose accuracy for most data sets, but some may require higher granularity in the discretization of features. Note that a higher number of bins
can increase the computation time significantly, in which case it may be beneficial to turn histograms off entirely.
The optimal values of the optional parameters is typically problem dependent. Cross-validation is typically used to tune options / hyperparameters.