Dealing with non-finite and missing values

[mattansb]

How should the various functions deal with posterior draws that have non-finite or missing values?

This can happen when computing ratios of posterior estimates (dividing by small values, or by 0)...

library(bayestestR)

x <- rnorm(10000, mean = 10)
hdi(x)
#> 95% HDI: [8.13, 11.98]
p_direction(x, null = 9)
#> Probability of Direction: 0.84
rope(x/100)
#> # Proportion of samples inside the ROPE [-0.10, 0.10]:
#> 
#> inside ROPE
#> -----------
#> 50.45 %

p_map(x)
#> MAP-based p-value: 0.00
point_estimate(x)
#> Point Estimate
#> 
#> Median |  Mean |  MAP
#> ---------------------
#> 9.99   | 10.00 | 9.92

Inf

x_inf <- c(x, Inf)
hdi(x_inf)
#> 95% HDI: [8.13, 11.98]
p_direction(x_inf, null = 9)
#> Probability of Direction: 0.84
rope(x_inf/100)
#> # Proportion of samples inside the ROPE [-0.10, 0.10]:
#> 
#> inside ROPE
#> -----------
#> 50.45 %

p_map(x_inf)
#> MAP-based p-value: 0.00
point_estimate(x_inf)
#> Point Estimate
#> 
#> Median | Mean |  MAP
#> --------------------
#> 9.99   |  Inf | 6.24

NA

x_na <- c(x, NA)
hdi(x_na)
#> 95% HDI: [8.13, 11.98]
p_direction(x_na, null = 9)
#> Probability of Direction: 0.84
rope(x_na/100)
#> # Proportion of samples inside the ROPE [-0.10, 0.10]:
#> 
#> inside ROPE
#> -----------
#> 50.46 %

p_map(x_na)
#> MAP-based p-value: 0.00
point_estimate(x_na)
#> Point Estimate
#> 
#> Median | Mean |  MAP
#> --------------------
#>        |      | 9.92

^{Created on 2022-12-07 with reprex v2.0.2}