Load an adaptive session from disk.
Details
Restores a persisted Adaptive state and revalidates basic invariants such
as schema version, required state fields, and index ranges in
step_log. If per-refit item logs are found on disk, they are loaded
into state$item_log and persistence is marked as enabled. Resume uses
strict schema validation for canonical logs; incompatible saved schemas abort
with explicit errors.
Legacy sessions without a saved predictive mode migrate to cold when
no predictive prior exists, and btl_only otherwise. An absent pairing
strategy migrates to hybrid. Saved TrueSkill values, the connected
bootstrap policy, frozen queue and its index, and round progress remain authoritative;
loading never recomputes predictions or initializes TrueSkill again.
Predictive-distribution sessions also record a versioned TrueSkill mapping, SD provenance, and integrity digests in state and session metadata. Loading validates their agreement without recomputing predictions or resetting ratings. Legacy sessions retain their existing migration rules and fixed-sigma policy.
metadata.rds records effective warm_start_mode and
pairing_strategy, plus bootstrap policy/version and the predictive
SHA-256 identity for session-level audit. Predictive queues are verified from
frozen inputs and the digest without repeating graph selection. Missing legacy
bootstrap policy fields retain shuffled behavior. summarize_adaptive() with
include_bootstrap = TRUE exposes the saved bootstrap audit. Direct step logs record
pairing_strategy, the presented A-over-B TrueSkill probability
p_ij, and target_distance (missing for random pairing). Predictive
vectors and provenance are retained once in state$predictive_prior.
See also
save_adaptive_session(), validate_session_dir(), adaptive_rank_resume()
Other adaptive persistence:
save_adaptive_session(),
validate_session_dir()
Examples
dir <- tempfile("pwllm-session-")
state <- adaptive_rank_start(c("a", "b", "c"), seed = 1)
save_adaptive_session(state, dir, overwrite = TRUE)
restored <- load_adaptive_session(dir)
summarize_adaptive(restored)
#> # A tibble: 1 × 6
#> n_items steps_attempted committed_pairs n_refits last_stop_decision
#> <int> <int> <int> <int> <lgl>
#> 1 3 0 0 0 FALSE
#> # ℹ 1 more variable: last_stop_reason <chr>