select: Approximate Best Subset Maximum Binary Prediction Rule...

In cykbennie/prescience-r: Approximate Best Subset Maximum Binary Prediction Rule (PRESCIENCE)

Description

select employs the Approximate Best Subset Maximum Binary Prediction Rule (PRESCIENCE) on the given variable selection problem.

Usage

select(formula, data, nfoc, q, bound, beta0 = 1, tol = NULL,
  warmstart = TRUE, tau = 1.5, mio = 1, tlim = 86400)

Arguments

formula	an object of class formula of the format: (binary dependent variable) ~ (normalized focus variable) + (remaining focus variables) + (auxiliary variables).
data	a data frame containing the variables in the model.
nfoc	integer. The number of focus variable(s) excluding the intercept.
q	integer. The cardinality constraint for the covariate selection.
bound	numeric. The maximum absolute value of the bounds for all variables.
beta0	integer. The coefficient taking value either 1 or -1 to normalize the scale for the first focus variable.
tol	numeric. Tolerance level. If NULL, use the default tolerance level.
warmstart	logical. If TRUE, use the warm start strategy.
tau	the tuning parameter for enlarging the estimated bounds.
mio	integer. 1 for MIO method 1 and 2 for method 2 in the paper.
tlim	time limit (in seconds) specified for the MIO solver.

Value

a list with 7 elements:

tolerance	tolerance level
status	optimization status
score	Gurobi score
gap	the MIO optimization gap value in case of early termination (0 if optimal solution is found within the time limit)
rtime	time used by the MIO solver in the estimation procedure
ncount	the number of Branch-and-bound nodes
bhat	maximum score estimates for coefficients

Author(s)

Yankang (Bennie) Chen <[e-mail]>

References

Best Subset Binary Prediction by Le-Yu Chen and Sokbae Lee (2018). https://arxiv.org/abs/1610.02738

Examples

results <- select(auto ~ dcost + cars + dovtt + divtt,
data = transportation, nfoc = 1, q = 1, bound = 10)

summary(results)
coef(results)

cykbennie/prescience-r documentation built on May 10, 2019, 3:16 a.m.