dx_dens: Density based numerical identifiability diagnostics

In alexpkeil1/vibr: Variable Importance in Black-Box Regression

Density based numerical identifiability diagnostics

Description

Give a numerical (but cruder) version of the diagnostics in plotshift_dens, where one can track the change in estimated exposure mass/density following a stochastic intervention on exposure.

Usage

dx_dens(vibr_fit, Acol = 1, delta = 0.01, quantiles = c(0, 0.1, 0.9, 1))

Arguments

vibr_fit	a fit from varimp
Acol	(integer) which column of predictors in call to varimp to diagnose
delta	(numeric, default=0.01) change in each column of predictors in call to varimp corresponding to stochastic intervention
quantiles	(numeric vector, default=c(0, 0.1, 0.9, 1)) cutpoints in the closed interval [0,1] that correspond to quantiles of the estimated density of observed values of a predictor. The length of this vector determines the size of the table. Using values close to 0 or 1 allows one to track whether "intervened" predictors are pushed toward the extreme of the estimated predictor density, which could indicate lack of support for the scale of the implied intervention (e.g. delta is too big).

Examples

## Not run: 
data(metals, package="qgcomp")
# subset of predictors
XYlist = list(X=metals[,1:4], Y=metals$y)
Y_learners = .default_continuous_learners()
Xbinary_learners = list(Lrnr_stepwise$new(name="SW"))
Xdensity_learners = .default_density_learners(n_bins=c(10))
set.seed(1231)
vi_ipw <- varimp(X=XYlist$X,Y=XYlist$Y, delta=0.1, Y_learners = Y_learners,
       Xdensity_learners=Xdensity_learners[1:2], Xbinary_learners=Xbinary_learners,
       estimator="IPW")
dx_dens(vi_ipw, Acol=1, delta=0.01)
# this shows that most observations keep a similar density range
dx_dens(vi_ipw, Acol=1, delta=0.01, quantiles=c(0, 0.1, 0.3, 0.7, 0.9, 1))

## End(Not run)

alexpkeil1/vibr documentation built on Sept. 13, 2023, 3:20 a.m.