Computes a given histogram distance metric to a distribution

Generally this function is used to estimate the parameters of a distribution by minimizing the given metric. The first argument is a vector of parameters that should match up with the arguments for dist. For example, params = c(0, 1) with dist = 'norm' refers to mean 0 and sd 1. This is so that we can use metric.histogram.dist in DEoptim::DEoptim which requires the parameters to optimize to be the first argument.

Usage

metric.histogram.dist(
  params,
  x,
  interval_start,
  interval_end,
  dist = c("norm", "gamma", "gamma_flip"),
  metric_func = ls(pattern = "^histogram[.].*", envir = getNamespace("HistogramZoo")),
  truncated = FALSE
)

Arguments

params

vector of parameters that should match up with the arguments for dist.

x

counts of the histogram

interval_start

left endpoint of the histogram

interval_end

right endpoint of the histogram

dist

one of norm, gamma or gamma_flip

metric_func

one of histogram.chisq, histogram.intersection, histogram.jaccard, histogram.ks, histogram.mse

truncated

Should distributions be truncated to the end points of the histogram?

Value

scalar results from comparing the distribution, with given parameters to observed histogram

Examples

x.hist <- observations_to_histogram(rnorm(1000))

args <- list(
  x = x.hist$histogram_data,
  interval_start = x.hist$interval_start,
  interval_end = x.hist$interval_end,
  dist = 'norm',
  metric_func = 'histogram.intersection'
  )

do.call(
  metric.histogram.dist,
  c(args, list(params = c(0, 1)))
  )
#> [1] 0.02419275

do.call(
  metric.histogram.dist,
  c(args, list(params = c(1, 1.5)))
  )
#> [1] 0.3569778