expected_learning: Expected learning

In lilymedina/gbiqqtools: Tools for generalized BIQQ

Description

Expected reduction in variance from one step data collection strategy

Usage

expected_learning(model, query, strategy = NULL, given = NULL,
  parameters = NULL)

Arguments

model	A model
query	A query as a character string, for example 'Y[X=1]>Y[X=0]'
strategy	A set of variables to be sought
given	A conditioning set as a character string that evaluates to a logical, for example 'Y==1'
parameters	a parameter vector

Examples

# Reduction in variance given monotonic X -> M1 -> M2 -> Y model
library(dplyr)
model <- make_model("X -> M1 -> M2 -> Y") %>%
  set_restrictions(node_restrict = list(M1 = "10", M2 = "10", Y = "10")) %>%
  set_priors() %>%
  set_parameters(type = 'flat')
el <- expected_learning(model, query = "Y[X=1]>Y[X=0]",
                  strategy = c("X", "M2"), given = "Y==1")
attr(el, "results_table")
el2 <- expected_learning(model, query = "Y[X=1]>Y[X=0]",
                  strategy = c("M1"),
                  given = "Y==1 & X==1 & M2==1")
attr(el2, "results_table")

# No strategy
expected_learning(model, query = "Y[X=1]>Y[X=0]")

# No givens
expected_learning(model, query = "Y[X=1]>Y[X=0]",
strategy = c("M1"))
expected_learning(model, query = "Y[X=1]>Y[X=0]",
strategy = c("M1"), given = "Y==1")

library(dplyr)
model <-  make_model("S -> C -> Y <- R <- X; X -> C -> R") %>%
set_restrictions(node_restrict =
list(C = "C1110", R = "R0001", Y = "Y0001"),
keep = TRUE)

expected_learning(model,
query = list(COE = "(Y[S=0] > Y[S=1])"),
strategy = "C", given = "Y==1 & S==0")

expected_learning(model,
query = list(COE = "(Y[X=1] > Y[X=0])"),
strategy = "S", given = "X==0 & Y==0")

lilymedina/gbiqqtools documentation built on Nov. 4, 2019, 4:32 p.m.