tests/testthat/test-distribs.R

In USGS-R/streamMetabolizer: Models for Estimating Aquatic Photosynthesis and Respiration

context('distribs')
# this file is really just to help me wrap my head around parameter
# distributions and parameter scaling.

library(ggplot2)

# variable description convention in notes:
# stan.and.SM.arg = symbol = r.arg

explore_scaling <- function() {
  #skip('tests are for developer understanding, not code performance')

  # location = mu = meanlog; scale = sigma = sdlog
  location <- 0
  scale <- 1
  n <- 10000

  # desired
  desired <- rlnorm(n, meanlog=location, sdlog=scale)
  expect_equal(mean(log(desired)), location, tol=0.5)
  expect_equal(sd(log(desired)), scale, tol=0.2)
  expect_gt(min(desired), 0)
  expect_equal(min(desired), 0, tol=0.1)
  ggplot(tibble::tibble(desired), aes(x=log(desired))) + geom_histogram()
  ggplot(tibble::tibble(desired), aes(x=desired)) + geom_histogram() + xlim(-1, 5)

  # scaled
  scaled <- rnorm(n, mean=0, sd=1)
  ggplot(tibble::tibble(scaled), aes(x=scaled)) + geom_histogram()
  final <- exp(location)*(exp(scaled)^scale)
  ggplot(tibble::tibble(final), aes(x=log(final))) + geom_histogram()
  ggplot(tibble::tibble(final), aes(x=final)) + geom_histogram() + xlim(-5, 100)

  # comparison
  comp <- dplyr::bind_rows(
    tibble::tibble(value=desired, type='desired'),
    tibble::tibble(value=scaled, type='scaled'),
    tibble::tibble(value=final, type='final'))
  ggplot(comp) + geom_histogram(aes(x=log(value), fill=type), alpha=0.5, binwidth=1, position='identity')
  ggplot(dplyr::filter(comp, type != 'scaled')) + geom_histogram(aes(x=value, fill=type), alpha=0.5, binwidth=1, position='identity') + xlim(-5, 100)
}