From 0cdede9c6bcdb0b66c6de4a9f76a7c681b0eca45 Mon Sep 17 00:00:00 2001 From: Claude Date: Fri, 11 Sep 2026 02:25:05 +0000 Subject: [PATCH] wiki(methods): ParallelEPOptimiser removed (PyAutoFit#1611); EP parallelism is JAX vectorisation Co-Authored-By: Claude Fable 5.1 Claude-Session: https://claude.ai/code/session_01FU8EkdahFWmrL6pewkHZR3 --- wiki/methods/concepts/expectation-propagation.md | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/wiki/methods/concepts/expectation-propagation.md b/wiki/methods/concepts/expectation-propagation.md index 052b62f..474e959 100644 --- a/wiki/methods/concepts/expectation-propagation.md +++ b/wiki/methods/concepts/expectation-propagation.md @@ -156,9 +156,12 @@ is an open design question (EP review Phase 5). LogGaussian `with_limits` crash, Gamma `from_mode` variance inversely wrong, negative-sigma unchecked, Beta clamp no-op, etc. Decisions hub: PyAutoFit #1331. -- `ParallelEPOptimiser` recomputes all cavities from the same stale mean - field per sweep (standard parallel-EP semantics, but different results - from serial) and crashes at completion when `paths=None`. (#1332 F3) +- `ParallelEPOptimiser` (EP's own process pool across factors) recomputed + all cavities from the same stale mean field per sweep and crashed at + completion when `paths=None` (#1332 F3). Removed 2026-09-11 (PyAutoFit + #1611): EP never combines with Python multiprocessing (ruling 2026-09-09, + PyAutoFit #1608) — factor searches run serially and parallelism comes from + a JAX-vectorised likelihood. ## Reading