Balancing Walk Design with Restarts

The Balancing Walk Design with Restarts. At each step, it adjusts randomization probabilities to ensure that imbalance tends towards zero.

Public fields

q

Target marginal probability of treatment.

intercept

Whether to add an intercept term.

delta

Probability of failure.

N

Total number of points.

D

Dimension of the data.

value_plus

Value added to imbalance when treatment is 1.

value_minus

Value added to imbalance when treatment is 0.

phi

Robustness parameter.

alpha

Normalizing constant (threshold).

w_i

Current imbalance vector.

iterations

Current iteration count.

Active bindings

definition

Dictionary of definition parameters.

state

Dictionary of current state.

Methods

Public methods

• BWD$new()

• BWD$set_alpha()

• BWD$process_x()

• BWD$update_imbalance()

• BWD$assign_next()

• BWD$replay_assignment()

• BWD$assign_all()

• BWD$update_state()

• BWD$reset()

• BWD$clone()

---

Method new()

Initialize the BWD balancer.

Usage

BWD$new(N, D, delta = 0.05, q = 0.5, intercept = TRUE, phi = 1)

Arguments

N

Total number of points.

D

Dimension of the data.

delta

Probability of failure (default 0.05).

q

Target marginal probability of treatment (default 0.5).

intercept

Whether to add an intercept term (default TRUE).

phi

Robustness parameter (default 1).

---

Method set_alpha()

Set normalizing constant for remaining N units.

Usage

BWD$set_alpha(N)

Arguments

N

Number of units remaining.

---

Method process_x()

Internal helper to process the input vector x. Returns the vector with intercept if applicable.

Usage

BWD$process_x(x)

Arguments

x

Covariate profile vector.

---

Method update_imbalance()

Manually update internal state based on an assignment. Used during Replay/Event Sourcing.

Usage

BWD$update_imbalance(x_proc, assignment)

Arguments

x_proc

Processed covariate vector (with intercept).

assignment

The treatment assignment (0 or 1).

---

Method assign_next()

Assign treatment to the next point.

Usage

BWD$assign_next(x)

Arguments

x

Covariate profile vector.

Returns

Treatment assignment (0 or 1).

---

Method replay_assignment()

Replay a historical assignment, respecting Restart logic.

Usage

BWD$replay_assignment(x, assignment)

Arguments

x

Covariate profile vector.

assignment

The treatment assignment (0 or 1).

---

Method assign_all()

Assign all points in a matrix (offline setting).

Usage

BWD$assign_all(X)

Arguments

X

Matrix of covariate profiles (N x D).

Returns

Vector of treatment assignments.

---

Method update_state()

Update the internal state of the balancer.

Usage

BWD$update_state(w_i, iterations, alpha = NULL)

Arguments

w_i

Current imbalance vector.

iterations

Current iteration count.

alpha

Optional. The current threshold value.

---

Method reset()

Reset the balancer to initial state.

Usage

BWD$reset()

---

Method clone()

The objects of this class are cloneable with this method.

Usage

BWD$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.