plot_hdpglm: Plot posterior distributions
In hdpGLM: Hierarchical Dirichlet Process Generalized Linear Models

plot_hdpglm

R Documentation

Plot posterior distributions

Description

this function creates a plot with two grids. One is the grid with posterior expectation of betas as function of context-level covariates. The other is the posterior distribution of tau

Usage

plot_hdpglm(
  samples,
  X = NULL,
  W = NULL,
  ncol.taus = 1,
  ncol.betas = NULL,
  ncol.w = NULL,
  nrow.w = NULL,
  smooth.line = FALSE,
  pred.pexp.beta = FALSE,
  title.tau = NULL,
  true.tau = NULL,
  title.beta = NULL,
  tau.x.axis.size = 1.1,
  tau.y.axis.size = 1.1,
  tau.title.size = 1.2,
  tau.panel.title.size = 1.4,
  tau.legend.size = 1,
  beta.x.axis.size = 1.1,
  beta.y.axis.size = 1.1,
  beta.title.size = 1.2,
  beta.panel.title.size = 1.4,
  beta.legend.size = 1,
  tau.xlab = NULL
)

Arguments

`samples`	an output of the function `hdpGLM`
`X`	a string vector with the name of the first-level covariates whose associated tau should be displayed
`W`	a string vector with the name of the context-level covariate(s) whose linear effect will be displayed. If `NULL`, the linear effect tau of all context-level covariates are displayed. Note: the context-level covariate must have been included in the estimation of the model.
`ncol.taus`	integer with the number of columns of the grid containing the posterior distribution of tau
`ncol.betas`	integer with the number of columns of the posterior expectation of betas as function of context-level features
`ncol.w`	integer with the number of columns to use to display the different context-level covariates
`nrow.w`	integer with the number of rows to use to display the different context-level covariates
`smooth.line`	boolean, if `TRUE` the plot will display a regression line representing the regression of the posterior expectation of the linear coefficients betas on the context-level covariates. Default `FALSE`
`pred.pexp.beta`	boolean, if `TRUE` the plots will display a line with the predicted posterior expectation of betas obtained using the posterior expectation of taus, the linear coefficients of the expectation of beta
`title.tau`	string, the title for the posterior distribution of the context effects
`true.tau`	a `data.frame` with four columns. The first must be named `w` and it indicates the index of each context-level covariate, starting with 0 for the intercept term. The second column named `beta` must contain the indexes of the betas of individual-level covariates, starting with 0 for the intercept term. The third column named `Parameter` must be named `tau<w><beta>`, where `w` and `beta` must be the actual values displayed in the columns `w` and `beta`. Finally, it must have a column named `True` with the true value of the parameter.
`title.beta`	string, the title for the posterior expectation of beta as function of context-level covariate
`tau.x.axis.size`	numeric, relative size of the x-axis of the plot with tau
`tau.y.axis.size`	numeric, relative size of the y-axis of the plot with tau
`tau.title.size`	numeric, relative size of the title of the plot with tau
`tau.panel.title.size`	numeric, relative size of the title of the panels of the plot with tau
`tau.legend.size`	numeric, relative size of the legend of the plot with tau
`beta.x.axis.size`	numeric, relative size of the x-axis of the plot with beta
`beta.y.axis.size`	numeric, relative size of the y-axis of the plot with beta
`beta.title.size`	numeric, relative size of the title of the plot with beta
`beta.panel.title.size`	numeric, relative size of the title of the panels of the plot with beta
`beta.legend.size`	numeric, relative size of the legend of the plot with beta
`tau.xlab`	string, the label of the x-axis for the plot with tau

Examples


library(magrittr)
# Note: this example is just for illustration. MCMC iterations are very reduced
set.seed(10)
n = 20
data.context1 = tibble::tibble(x1 = rnorm(n, -3),
                                   x2 = rnorm(n,  3),
                                   z  = sample(1:3, n, replace=TRUE),
                                   y  =I(z==1) * (3 + 4*x1 - x2 + rnorm(n)) +
                                       I(z==2) * (3 + 2*x1 + x2 + rnorm(n)) +
                                       I(z==3) * (3 - 4*x1 - x2 + rnorm(n)) ,
                                   w = 20
                                   ) 
data.context2 = tibble::tibble(x1 = rnorm(n, -3),
                                   x2 = rnorm(n,  3),
                                   z  = sample(1:2, n, replace=TRUE),
                                   y  =I(z==1) * (1 + 3*x1 - 2*x2 + rnorm(n)) +
                                       I(z==2) * (1 - 2*x1 +   x2 + rnorm(n)),
                                   w = 10
                                   ) 
data = data.context1 %>%
    dplyr::bind_rows(data.context2)

## estimation
mcmc    = list(burn.in=1, n.iter=50)
samples = hdpGLM(y ~ x1 + x2, y ~ w, data=data, mcmc=mcmc, n.display=1)

plot_hdpglm(samples)
plot_hdpglm(samples, ncol.taus=2, ncol.betas=2, X='x1')
plot_hdpglm(samples, ncol.taus=2, ncol.betas=2, X='x1', ncol.w=2, nrow.w=1,
            pred.pexp.beta=TRUE,smooth.line=TRUE )

hdpGLM documentation built on Oct. 13, 2023, 1:17 a.m.