Enter simulation workflow mode. Use when configuring, running, debugging, or analyzing simulation experiments.
You are helping configure, run, debug, or analyze simulation experiments for the SuSiNE benchmarking study.
@refs/susine_study_plan_high_level.md @refs/susie_susine_background.md
Read these when the user asks about why the harness is structured a certain way, when adding new scenarios or use cases, or when making architectural changes to the simulation pipeline. They contain the study hypotheses (H1-H6), failure-mode characterization framework, and the phased study structure that drives what the harness needs to support.
1. Configure --> make_job_config() or run_control_workbook.Rmd
2. Write --> write_job_artifacts() produces job_config.json, run_table.csv, .slurm
3. Submit --> sbatch output/slurm_scripts/<job>.slurm
4. Execute --> Each SLURM array task calls inst/scripts/run_task.R -> run_task()
5. Collect --> aggregate_staging_outputs() combines flush files
6. Visualize --> visualize_results_workbook_paper_exhibits.Rmd
| Parameter | Description | Example values |
|---|---|---|
use_case_ids | Model configs from use_case_catalog() | c("a_i", "b_ii", "c_ii") |
L_grid | Number of single effects | 5, 10, c(5, 10) |
p_star_grid | True causal SNP count | c(1, 2, 3, 4, 5, 10, 20) |
y_noise_grid | Noise fraction (0=no noise) | c(0.05, 0.1, 0.2, 0.4) |
prior_quality | Annotation settings | prior_quality_grid(...) |
seeds | RNG seeds for phenotype sim | 1:3 |
data_scenarios | Genotype source | "simulation_n3" |
grid_mode | How grids combine | "full", "minimal", "intersect" |
sigma_0_2_scalars | Prior variance multipliers | c("0.1", "1/L") |
annotation_scales | c-grid for functional mu | c(0, 0.5, 1, 2, 5) |
aggregation_methods | BMA strategies | c("softmax_elbo", "mean", "max_elbo") |
restart_settings | Random init config | list(n_inits = 5, alpha_concentration = 1) |
output/
temp/<job>/ # Temporary config (overwritten on re-create)
job_config.json
run_table.csv
dataset_bundles.csv
use_cases.csv
run_history/<job>/<parent>/ # Immutable config snapshot
slurm_output/<job>/<parent>/ # Per-task flush files
task-001/
flush-000_model_metrics.csv
flush-000_confusion_bins.csv
flush-000_snps.parquet
aggregated/ # Post-collection output
model_metrics.csv
confusion_bins.csv
snps_dataset/ # Partitioned by use_case_id
slurm_scripts/<job>.slurm
slurm_prints/<job>/<parent>/ # SLURM stdout/stderr