montecarlo opなし → table(引数だけで決まる op —— 画像やデータの入力を取らない)import fullseye as fs; fs.ledger.mc_sequential_mean(sampler='pi', tol=0.01, rel_tol=0.0, confidence=0.95, batch=4096, max_samples=4000000, seed=0, on_fail='raise', history_stride=1) -> 'dict' (実装を直接呼ぶなら import converge; converge.mc_sequential_mean(sampler='pi', tol=0.01, rel_tol=0.0, confidence=0.95, batch=4096, max_samples=4000000, seed=0, on_fail='raise', history_stride=1) -> 'dict'、台帳から引くなら opsconverge.get("mc_sequential_mean"))Sequential Monte Carlo mean: add samples until the confidence interval is narrower than the tolerance.
sampler(rng, n) returns n i.i.d. samples (a 1-D array) — or the built-in "pi"
(4 * [u^2 + v^2 <= 1], mean pi). Batch b is drawn from default_rng((seed, b)), so the
result is reproducible and does not depend on how many batches came before.
Stops at the first batch boundary, from the second batch on, where the half-width
z * s / sqrt(n) <= tol + rel_tol * |mean| (z from confidence; s the running
sample standard deviation). This is the fixed-width sequential interval of Chow & Robbins
1965; never stopping inside the first two batches guards against an early, accidentally
small variance (raise batch to guard more) — the actual coverage is checked by the
gates, not assumed.
Returns mean, se, ci (lower, upper), half_width, n_samples (used),
converged and history (n, mean, se per batch). Refusal: max_samples reached
first raises NotConvergedError (on_fail="record" returns the record with
mean = None and the running values under last_mean / last_se).
table を入力に取れる)—
montecarlo)Provenance: converge.py — CONVERGE operator registry. この per-op ノートは tools/opdocs.py md が自動生成(手編集しない)。
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.