Displays information about the inference modes available in tulpaRatio, including their tiers, guarantees, and appropriate use cases.
Examples
inference_mode_info()
#> tulpaRatio Inference Modes
#> ========================
#>
#> Modes encode EPISTEMIC GUARANTEES, not just runtime characteristics.
#>
#> TIER 1: EXACT
#> Asymptotically correct posterior inference
#> Guarantee: Credible intervals interpretable as posterior uncertainty
#> Backends: hmc, ess, pg, gibbs, sghmc, sgld
#> Note: Reference standard
#>
#> TIER 2: STRUCTURED
#> Accurate inference conditional on structural assumptions
#> Guarantee: Correct if model meets structural assumptions
#> Backends: laplace
#> Note: Controlled approximation, not heuristics
#>
#> TIER 3: OPTIMIZED
#> Optimization-based approximate inference
#> Guarantee: No general correctness guarantee
#> Backends: vi
#> Note: Uncertainty often underestimated; requires explicit opt-in
#>
#> AUTO MODE
#> Selects between Tier 1 and Tier 2 only.
#> NEVER silently chooses Tier 3 (Optimized).
#> Contract: 'Use the most reliable method expected to finish.'
#>
#> Usage (by tier):
#> tratio(..., mode = 'auto') # Tier 1 or 2 (default)
#> tratio(..., mode = 'exact') # Tier 1
#> tratio(..., mode = 'structured') # Tier 2
#> tratio(..., mode = 'optimized') # Tier 3 (explicit opt-in)
#>
#> Usage (by backend):
#> tratio(..., mode = 'hmc') # HMC/NUTS (Tier 1)
#> tratio(..., mode = 'ess') # Elliptical Slice Sampling (Tier 1)
#> tratio(..., mode = 'sghmc') # Stochastic Gradient HMC (Tier 1, large N)
#> tratio(..., mode = 'sgld') # Stochastic Gradient Langevin (Tier 1, large N)
#> tratio(..., mode = 'laplace') # Laplace approximation (Tier 2)
#> tratio(..., mode = 'vi') # Variational Inference (Tier 3)