FLAME_bit: Bit Vectors Implementation

In chiarui424/FLAME: A Fast Large-Scale Almost Matching Exactly Approach to Causal Inference

Description

FLAME_bit applies FLAME matching algorithm based on bit vectors. The required arguments include (1) data and (2) holdout. The default model for Match Quality is set to Ridge regression with 0.1 regularization parameter.

Usage

FLAME_bit(data, holdout, tradeoff = 0.1, compute_var = FALSE,
  PE_function = NULL, model = NULL, ridge_reg = NULL,
  lasso_reg = NULL)

Arguments

data	input data
holdout	holdout training data
tradeoff	Match Quality tradeoff parameter (optional, default = 0.1)
compute_var	variance indicator (optional, default = FALSE)
PE_function	user defined function to compute predictive error (optional)
model	Linear, Ridge, or Lasso (optional)
ridge_reg	L2 regularization parameter if model = Ridge (optional)
lasso_reg	L1 regularization parameter if model = Lasso (optional)

Value

(1) list of covariates FLAME performs matching at each iteration, (2) Sizes, conditional average treatment effects (CATEs), and variance (if compute_var = TRUE) of matches at each iteration, (3) match quality at each iteration, and (4) the original data with additional column *matched*, indicating the number of covariates each unit is matched on. If a unit is never matched, then *matched* will be 0.

Examples

data(toy_data)
FLAME_bit(data = toy_data, holdout = toy_data)

chiarui424/FLAME documentation built on Sept. 16, 2019, 8:43 a.m.