Skip to contents

Forest plot of ADEMP Monte Carlo summaries

Source: R/04_02_benchmark_visualisation.R
plot_mc_forest.Rd

Dot-and-whisker display of bias, RMSE, MAE, coverage, mean interval width, SE/SD ratio, and optimiser failure rate by algorithm and cell type (Allen et al. 2019) . Coverage whiskers are the Wilson interval already stored on monte_carlo (coverage_mc_interval(); (Wilson 1927) ): they are intervals for the coverage rate, not for \(p_j\). Bias is referenced at 0; coverage at 0.95; SE/SD (se_sd_ratio, mean model SE / empirical SD) at 1. Interior Wald SEs come from vcov_ilr_delta(). Pairwise algorithm contrasts (MAE differences versus a reference solver on the same Monte Carlo replicates) can be read from the raincloud of paired errors; they do not need a second bootstrap.

Usage

plot_mc_forest(
  benchmark,
  facet_rows = NULL,
  facet_cols = NULL,
  metrics = c("bias", "rmse", "mae", "coverage", "mean_interval_width", "se_sd_ratio",
    "failure_rate")
)

Arguments

benchmark

List from run_simulation_benchmark(), or a table from pivot_mc_estimates().

facet_rows, facet_cols

Optional column names for ggplot2::facet_grid() rows and columns.

metrics

Character vector of summaries to display.

Value

A ggplot object.

References

Allen M, Poggiali D, Whitaker K, Marshall TR, van Langen J, Kievit RA (2019). “Raincloud Plots: A Multi-Platform Tool for Robust Data Visualization.” Wellcome Open Research, 4, 63. ISSN 2398-502X. doi:10.12688/wellcomeopenres.15191.2 . https://wellcomeopenresearch.org/articles/4-63.

Wilson EB (1927). “Probable Inference, the Law of Succession, and Statistical Inference.” Journal of the American Statistical Association, 22(158), 209–212. ISSN 0162-1459. doi:10.2307/2276774 .

See also

plot_mc_raincloud(), coverage_mc_interval(), vcov_ilr_delta(), theme_decovart_facets()

Examples

set.seed(1)
genes <- paste0("g", 1:2)
cts <- paste0("ct", 1:2)
mu <- matrix(c(20, 22, 22, 20), nrow = 2, dimnames = list(genes, cts))
Sigma <- array(
  c(1, 0, 0, 1, 1, 0, 0, 1),
  dim = c(2, 2, 2),
  dimnames = list(genes, genes, cts)
)
out <- run_simulation_benchmark(
  tibble::tibble(
    n_genes = 2L,
    cosine = 0,
    true_theta = list(list(p = c(0.5, 0.5), mu = mu, sigma = Sigma))
  ),
  deconvolution_functions = list(
    "nnls" = list(FUN = deconvolute_ratios_nnls)
  ),
  n = 4,
  cores = 1
)
if (requireNamespace("ggplot2", quietly = TRUE)) {
  plot_mc_forest(out, facet_cols = "cosine")
}