Skip to contents

Benchmark diagnostic agreement with a simulator-defined target

Source: R/benchmark_hrri.R
benchmark_hrri.Rd

Runs independent seeds and reports descriptive agreement, not held-out prediction, empirical validation or parameter identification.

Usage

benchmark_hrri(
  domains = c("soil", "plant", "micro"),
  missing = 0,
  noise = 0,
  n = 50L,
  seed_start = 1L,
  sim_args = NULL,
  verbose = TRUE,
  pipeline_args = list()
)

# S3 method for class 'hrri_benchmark'
print(x, ...)

Arguments

domains

Nonempty subset of soil, plant and micro.

missing

Fraction in [0, 1) of uniformly sampled cells removed (MCAR). This does not implement informative or MNAR missingness.

noise

Nonnegative Gaussian noise SD on each column's original scale. A common SD has different relative effects on differently scaled variables.

n

Positive integer number of independent simulations.

seed_start

First integer seed. The caller's RNG state is restored.

sim_args

Named simulator arguments, excluding seed.

verbose

Print progress messages.

pipeline_args

Named additional rri_pipeline arguments, excluding dat, soil, plant, micro and id; use this to justify orientations and weights.

x

An hrri_benchmark object.

...

Unused method arguments.

Value

An hrri_benchmark list with summary, seed_metrics, row_data, failures (seed and error), and settings including package/R versions. Summary RMSE and Bias compare the score directly with the chosen target. r_truth and rank_truth are pooled descriptive correlations; spread_association compares within-seed score and target SDs. None is interval coverage or predictive uncertainty. n_rows counts finite matched rows across all seeds.

Details

Soil and plant synthetic observations and log1p gene abundances feed the pipeline. Latent architecture and microbial activity states are not scoring inputs. The latent target is still a prescribed simulator composite; it is not an independently measured recovery outcome. Replicated rows within seeds are dependent. Compare methods using held-out seeds and independent process outcomes in a separate validation design.

See also

rri_pipeline, simulate_redox_holobiont, plot_hrri_benchmark

Examples

# \donttest{
b <- benchmark_hrri(domains = "soil", n = 2, missing = 0.1)
#> Warning: Excluding simulator-derived hidden columns from scoring: Cacc_EAC, Cacc_EDC, Cacc_total, Cacc_fraction, net_oxidative_balance, alpha_accept, alpha_donate, k_accept_h, k_donate_h
#> Warning: Unanchored latent axes have arbitrary signs; RRI is exploratory, not directionally validated resilience.
#> benchmark_hrri: 1/2 seeds attempted
#> Warning: Excluding simulator-derived hidden columns from scoring: Cacc_EAC, Cacc_EDC, Cacc_total, Cacc_fraction, net_oxidative_balance, alpha_accept, alpha_donate, k_accept_h, k_donate_h
#> Warning: Unanchored latent axes have arbitrary signs; RRI is exploratory, not directionally validated resilience.
#> benchmark_hrri: 2/2 seeds attempted
print(b)
#> <hrri_benchmark> descriptive agreement, not predictive validation
#>       RMSE        Bias    r_truth rank_truth n_rows spread_association
#>  0.3215426 -0.08773519 -0.3207642 -0.3211588    120                 NA
#>  mean_within_seed_sd n_requested n_seeds n_failed
#>            0.3054143           2       2        0
# }