paper-3PSM-and-excess-significance-test/simulations-power.R

In jepusto/metascore: Model-Based and Robust Score Tests for Meta-Regression

rm(list=ls())
source("paper-3PSM-and-excess-significance-test/simulation-functions.R")

source_obj <- ls()

#--------------------------------------------------------
# Simulation conditions
#--------------------------------------------------------
library(tidyverse)

set.seed(20190611)

design_factors <- list(
  studies = c(40, 80, 120),
  mean_effect = c(0.0, 0.2, 0.4, 0.8), 
  sd_effect = c(0.0, 0.2, 0.4),
  p_thresholds = .025, 
  p_RR = c(seq(0, 0.1, 0.02), seq(0.2, 0.9, 0.1))
)

lengths(design_factors)
prod(lengths(design_factors))

params <- 
  cross_df(design_factors) %>%
  mutate(
    reps = 8000,
    seed = round(runif(1) * 2^30) + row_number()
  ) %>%
  sample_frac()

params

#--------------------------------------------------------
# run simulations in parallel - furrr
#--------------------------------------------------------

library(future)
plan(multiprocess)
library(furrr)

tm <- system.time(
  results <- 
    params %>%
    mutate(
      res = future_pmap(., .f = possibly(runSim, NULL), 
                        n_sim = n_beta(20,100,1,1.5),
                        methods = c("FE","ML","REML","WLS"),
                        score_types = c("TES-norm","TES-binom","parametric","robust"),
                        LRT = TRUE, k_min = 2L)
    ) 
)

tm
cat("Projected time:", (8000 / unique(results$reps)) * (tm[[3]] / 60^2), "hours")

#--------------------------------------------------------
# run simulations in parallel - mdply
#--------------------------------------------------------

# library(Pusto)
# 
# cluster <- start_parallel(source_obj = source_obj, register = TRUE)
# 
# tm <- system.time(
#   results <- plyr::mdply(params, .f = runSim, 
#                          n_sim = n_beta(20,100,1,1.5), 
#                          methods = c("FE","ML","REML","WLS"),
#                          score_types = c("TES-norm","TES-binom","parametric","robust"),
#                          LRT = TRUE, k_min = 2L, 
#                          .parallel = TRUE)
# )
# 
# tm 
# parallel::stopCluster(cluster)


#--------------------------------------------------------
# Save results and details
#--------------------------------------------------------

session_info <- sessionInfo()
run_date <- date()

save(params, results, session_info, run_date, 
     file = "paper-3PSM-and-excess-significance-test/Power-results.Rdata")