Extends BWD to multiple treatments by constructing a binary tree where
each node balances between the treatment groups on the left and right.
Public fields
NTotal number of points.
DDimension of the data.
deltaProbability of failure.
interceptWhether to add an intercept term.
phiRobustness parameter.
qsTarget marginal probabilities for each treatment.
classesVector of class labels.
KNumber of treatment groups minus 1.
nodesList to store BWD objects or integers (leaves).
weightsList to store weights for tree construction.
Active bindings
definitionDictionary of definition parameters.
stateDictionary of current state.
Methods
Method new()
Initialize the MultiBWD balancer.
Usage
MultiBWD$new(N, D, delta = 0.05, q = 0.5, intercept = TRUE, phi = 1)
Arguments
NTotal number of points.
DDimension of the data.
deltaProbability of failure (default 0.05).
qTarget marginal probabilities. Can be a scalar (0.5 implied for 2 groups) or vector.
interceptWhether to add an intercept term (default TRUE).
phiRobustness parameter (default 1).
Method build_tree()
Internal helper to build the tree structure.
Method assign_next()
Assign treatment to the next point.
Arguments
xCovariate profile vector.
Returns
Treatment assignment (integer index).
Method assign_all()
Assign all points in a matrix (offline setting).
Arguments
XMatrix of covariate profiles (N x D).
Returns
Vector of treatment assignments.
Method update_state()
Update the internal state of the balancer.
Usage
MultiBWD$update_state(...)
Arguments
...Named arguments mapping node indices to state lists.
Method update_path()
Helper to manually update path for replay logic.
Usage
MultiBWD$update_path(x, final_assignment)
Arguments
xCovariate vector.
final_assignmentThe assigned class integer.
Method replay_assignment()
Replay assignment through the tree.
Usage
MultiBWD$replay_assignment(x, final_assignment)
Arguments
xCovariate profile vector.
final_assignmentThe treatment assignment (integer).
Method get_path_for_assignment()
BFS to find the path to a specific leaf class.
Usage
MultiBWD$get_path_for_assignment(target_class)
Arguments
target_classInteger class label.
Method reset()
Reset the balancer to initial state.
Method clone()
The objects of this class are cloneable with this method.
Usage
MultiBWD$clone(deep = FALSE)
Arguments
deepWhether to make a deep clone.