plot_pit: PIT histogram

In epiforecasts/scoringutils: Utilities for Scoring and Assessing Predictions

PIT histogram

Description

Make a simple histogram of the probability integral transformed values to visually check whether a uniform distribution seems likely.

Usage

plot_pit(pit, num_bins = "auto", breaks = NULL)

Arguments

pit	Either a vector with the PIT values, or a data.table as produced by get_pit().
num_bins	The number of bins in the PIT histogram, default is "auto". When num_bins == "auto", plot_pit() will either display 10 bins, or it will display a bin for each available quantile in case you passed in data in a quantile-based format. You can control the number of bins by supplying a number. This is fine for sample-based pit histograms, but may fail for quantile-based formats. In this case it is preferred to supply explicit breaks points using the breaks argument.
breaks	Numeric vector with the break points for the bins in the PIT histogram. This is preferred when creating a PIT histogram based on quantile-based data. Default is NULL and breaks will be determined by num_bins. If breaks is used, num_bins will be ignored.

Value

A ggplot object with a histogram of PIT values

Examples

# PIT histogram in vector based format
observed <- rnorm(30, mean = 1:30)
predicted <- replicate(200, rnorm(n = 30, mean = 1:30))
pit <- pit_sample(observed, predicted)
plot_pit(pit)

# quantile-based pit
pit <- get_pit(as_forecast(example_quantile), by = "model")
plot_pit(pit, breaks = seq(0.1, 1, 0.1))

# sample-based pit
pit <- get_pit(as_forecast(example_sample_discrete), by = "model")
plot_pit(pit)

epiforecasts/scoringutils documentation built on April 23, 2024, 4:56 p.m.