Computing Method Results • PublicationBiasBenchmark

This vignette explains how to compute and store results for a new publication bias correction method across all benchmark conditions and repetitions. This is a crucial step after implementing a new method (see Adding New Methods) to make it part of the benchmark comparisons. Once you compute and store the results, you can evaluate the method’s performance using the Computing Method Measures vignette, however, the package maintainers will update the precomputed measures upon your submission of the new method accompanied by the computed method results.

Overview

After implementing the three required functions for a new method (method.{METHOD_NAME}(), method_settings.{METHOD_NAME}(), and method_extra_columns.{METHOD_NAME}()), you need to:

Download presimulated datasets for each DGM
Apply your method to all condition-repetition combinations
Store the results in the appropriate directory structure
Document the session information for reproducibility

This process ensures that your method can be compared to existing methods using the precomputed performance measures.

Prerequisites

Before computing results, ensure that:

Your method is fully implemented with all three required functions
You have sufficient computational resources (the full benchmark may take hours to days depending on method complexity)
You have adequate disk space for storing results (typically several hundred MB per DGM)

Directory Structure

The benchmark uses the following directory structure for storing results:

resources/
├── {DGM_NAME}/
│   ├── data/
│   │   └── {condition_id}.csv          # Presimulated datasets
│   ├── results/
│   │   └── {METHOD}-{SETTING}.csv      # Method results
│   └── metadata/
│       ├── dgm-conditions.csv          # Condition parameters
│       ├── {METHOD}-{SETTING}-sessionInfo.txt
│       └── {METHOD}-{SETTING}-session.log

By default, this structure is created in the directory specified by PublicationBiasBenchmark.get_option("simulation_directory").

Computing Results: Step-by-Step Guide

Step 1: Set Up Your Environment

First, ensure you’re working in the correct directory and have the package loaded:

library(PublicationBiasBenchmark)

# Optional: Set custom directory for results storage
# PublicationBiasBenchmark.options(simulation_directory = "/path/to/storage")

# Verify the directory
data_folder <- PublicationBiasBenchmark.get_option("simulation_directory")
print(paste("Working directory:", data_folder))

Step 2: Define DGMs and Method Information

Specify which DGMs you want to compute results for and your method details:

# List of DGMs to evaluate
dgm_names <- c(
  "Stanley2017",
  "Alinaghi2018",
  "Bom2019",
  "Carter2019"
)

# Your method information
method_name    <- "myNewMethod"
method_setting <- "default"  # Or other setting name if you have multiple

Step 3: Download Presimulated Datasets

Download the presimulated datasets for all DGMs:

# Download datasets for all DGMs
for (dgm_name in dgm_names) {
  message("Downloading datasets for: ", dgm_name)
  download_dgm_datasets(dgm_name)
}

This step only needs to be done once. The datasets will be cached locally for subsequent use.

Step 4: Compute Results for Each DGM

Now, compute results for your method across all conditions and repetitions. The code bellow is a template that you can adapt as needed. You will most likely adapt this code to run parallel/using a job scheduler on a cluster.

# Set seed for reproducibility
set.seed(1)

# Process each DGM
for (dgm_name in dgm_names) {
  
  message("Processing DGM: ", dgm_name)
  
  # Get condition information
  conditions <- dgm_conditions(dgm_name)
  message("Number of conditions: ", nrow(conditions))
  
  # Container to store all results for this DGM
  all_results <- list()
  
  # Process each condition
  for (condition_id in conditions$condition_id) {
    
    message("  Condition ", condition_id, " / ", nrow(conditions))
    
    # Retrieve all repetitions for this condition
    condition_datasets <- retrieve_dgm_dataset(
      dgm_name = dgm_name,
      condition_id = condition_id,
      repetition_id = NULL  # NULL retrieves all repetitions
    )
    
    # Get unique repetition IDs
    repetition_ids <- unique(condition_datasets$repetition_id)
    message("    Repetitions: ", length(repetition_ids))
    
    # Compute results for each repetition
    condition_results <- list()
    for (repetition_id in repetition_ids) {
      
      # Extract data for this specific repetition
      repetition_data <- condition_datasets[
        condition_datasets$repetition_id == repetition_id, 
      ]
      
      # Apply your method (error handling is done internally)
      result <- run_method(
        method_name = method_name,
        data        = repetition_data,
        settings    = method_setting
      )
      
      # Attach metadata
      result$condition_id  <- condition_id
      result$repetition_id <- repetition_id
      
      condition_results[[repetition_id]] <- result
    }
    
    # Combine results for this condition
    all_results[[condition_id]] <- do.call(rbind, condition_results)
  }
  
  # Combine all results for this DGM
  dgm_results <- do.call(rbind, all_results)
  
  # Save results
  results_dir <- file.path(data_folder, dgm_name, "results")
  if (!dir.exists(results_dir)) {
    dir.create(results_dir, recursive = TRUE)
  }
  
  results_file <- file.path(
    results_dir, 
    paste0(method_name, "-", method_setting, ".csv")
  )
  
  write.csv(dgm_results, file = results_file, row.names = FALSE)
  message("Results saved to: ", results_file)
  
  # Save session information
  metadata_dir <- file.path(data_folder, dgm_name, "metadata")
  if (!dir.exists(metadata_dir)) {
    dir.create(metadata_dir, recursive = TRUE)
  }
  
  # sessionInfo() output
  sessioninfo_file <- file.path(
    metadata_dir,
    paste0(method_name, "-", method_setting, "-sessionInfo.txt")
  )
  writeLines(
    capture.output(sessionInfo()), 
    sessioninfo_file
  )
  
  # Detailed session info (using the sessioninfo package)
    session_log_file <- file.path(
      metadata_dir,
      paste0(method_name, "-", method_setting, "-session.log")
    )
    sessioninfo::session_info(to_file = session_log_file)
  
  message("Session info saved to: ", metadata_dir)
}

message("All computations completed!")

Contributing to the Package

The package maintainers will compute and update precomputed measures based on your computed results when you contribute a new method. If you want to contribute a new method:

Implement the method following the Adding New Methods guidelines
Compute results for all DGMs (desribed in this vignette)
Submit a pull request with your method implementation and results
Package maintainers will compute the measures and integrate them into the benchmark