ggdmcPrior (version 0.2.9.0)

BuildPrior: Build a Joint Prior Distribution

Description

BuildPrior sets up a joint distribution of the prior. Each model parameter is assigned one probability distribution. p0 and p1 refer to the first and second parameters. I use the convention of the 0-based index to work with the C++ and the Python sister package, 'pydmc' (coming soon). p0 must comes with parameter names.

Usage

BuildPrior(
  p0,
  p1,
  lower = rep(NA, length(p0)),
  upper = rep(NA, length(p0)),
  dists = rep("norm", length(p0)),
  log_p = rep(TRUE, length(p0)),
  types = c("tnorm", "beta", "gamma", "lnorm", "cauchy", "unif", "norm")
)

Value

a list of lists, where each sub-list contains the parameter for its prior definition. Each sub-list includes:

• p0: The first parameter of the distribution.

• p1: The second parameter of the distribution.

• lower: The lower bound of the distribution.

• upper: The upper bound of the distribution.

• dist: A numeric code representing the distribution type.

• log_p: Logical indicating whether probabilities are logged.

Arguments

p0

the first parameter of a distribution (e.g., mean, shape1, etc.).

p1

the second parameter of a distribution (e.g., sd, shape2, etc.).

lower

lower support (boundary). Default is NA, which will be converted to a real value or -Inf based on the distribution type.

upper

upper support (boundary). Default is NA, which will be converted to a real value or Inf based on the distribution type.

dists

a vector of character strings specifying the distribution type for each parameter. Valid types are: "tnorm", "beta", "gamma", "lnorm", "cauchy", "unif", and "norm". Default is "norm".

log_p

logical; if TRUE, probabilities are given as log(p). Default is TRUE.

types

available distribution types.

Details

Seven distributions are implemented:

1. Truncated normal distribution, where: p0 = mean, p1 = sd. When the lower and upper bounds are not provided, they are set to -Inf and Inf, rendering a normal distribution (see tnorm). Type name is "tnorm".

2. Beta distribution, where: p0 = shape1 and p1 = shape2 (see pbeta). Note the uniform distribution is a special case of the beta with p0 = 1 and p1 = 1. Type name is "beta".

3. Gamma distribution, where p0 = shape and p1 = scale (see pgamma). Note p1 is scale, not rate. Type name is "gamma".

4. Log-normal, where p0 = meanlog and p1 = sdlog (see plnorm). Type name is "lnorm".

5. Cauchy distribution, where p0 = location and p1 = scale (see pcauchy). Type name is "cauchy".

6. Uniform distribution, where p0 = lower and p1 = upper (see punif). Type name is "unif".

7. Normal distribution, where p0 = mean and p1 = sd (see pnorm). Type name is "norm".

Examples

# Using dbeta to represent a uniform distribution of bounds(0, 1)
x <- seq(-.1, 1.1, .001)
plot(x, dbeta(x, 1, 1),
    type = "l", ylab = "Density", xlab = "x",
    lwd = 2, cex.lab = 1.5, cex.axis = 2
)

## Create an S4 prior object
p_prior <- BuildPrior(
    p0 = c(A = 0.15, B = 0.45, mean_v = 2.25, sd_v = 0.15, t0 = 0.2),
    p1 = rep(0.1, 5),
    lower = rep(NA, 5),
    upper = rep(NA, 5),
    dist = rep("tnorm", 5),
    log_p = rep(NA, 5)
)

print_prior(p_prior)

# Use the beta distribution to create uniform densities
# lower and upper set the bounds. If lower is NA, it will be set to 0.
# If upper is NA, it will be set to 1.
p_prior <- BuildPrior(
    p0 = c(A = 1, B = 1, mean_v = 1, sd_v = 1, t0 = 1),
    p1 = rep(1, 5),
    lower = rep(0, 5),
    upper = rep(5, 5),
    dist = rep("beta", 5),
    log_p = rep(FALSE, 5)
)

p0 <- plot_prior(p_prior, font_size = 3.5, cex = 3.5)