plot_dependence {StratifiedMedicine}

Partial dependence plots: Single Variable (marginal effect) or heat map (2 to 3 variables).

Description

Partial dependence plots: Single Variable (marginal effect) or heat map (2 to 3 variables).

Usage

plot_dependence(object, X = NULL, target = NULL, vars, grid.data = NULL, ...)

Arguments

object

Fitted ple_train or PRISM object

X

input covariate space. Default=NULL.

target

Which patient-level estimate to target for PDP based plots. Default=NULL, which uses the estimated treatment difference.

vars

Variables to visualize (ex: c("var1", "var2", "var3)). If no grid.data provided, defaults to using seq(min(var), max(var)) for each continuous variables. For categorical, uses all categories.

grid.data

Input grid of values for 2-3 covariates (if 3, last variable cannot be continuous). This is required for type="heatmap". Default=NULL.

...

Additional arguments (currently ignored).

Value

Plot (ggplot2) object

References

• Friedman, J. Greedy function approximation: A gradient boosting machine. Annals of statistics (2001): 1189-1232

• Zhao, Qingyuan, and Trevor Hastie. Causal interpretations of black-box models. Journal of Business & Economic Statistics, to appear. (2017).

Examples

library(StratifiedMedicine)
## Continuous ##
dat_ctns = generate_subgrp_data(family="gaussian")
Y = dat_ctns$Y
X = dat_ctns$X
A = dat_ctns$A


# Fit through ple_train wrapper #
mod = ple_train(Y=Y, A=A, X=X, Xtest=X, ple="ranger", meta="X-learner")
plot_dependence(mod, X=X, vars="X1")