9923c73 · atomspace-gpu

atomspace-gpu commits / 9923c73

Step 2 (partial): gpu-sections.cl audited — data-model matches opencog, 6/6 on GPU

Audit of gpu-sections.cl: its data model is exactly opencog's Section/
ConnectorSeq/Connector (section = (word, sorted-connector-set, count);
connector = (partner_word, +/-dir)). test-sections.c verifies against
hand-computed ground truth (specified disjuncts + CPU hashes, dedup
count=2.0, no cross-sentence-boundary sections) and builds+runs on the
RTX 2070 -> 6/6 PASS.

OPEN: disjunct is hashed (FNV-1a over sorted connectors), not symbolic,
so the opencog data-tie is STRUCTURAL (match (word, connector-multiset,
count) against opencog's pseudo-cset extraction) and needs the opencog
MST pipeline (word-pair MI -> MST -> cset) -- a sub-project, deferred.
Sections now sit at MI's pre-1b bar (data-model + hand-verified + runs
on GPU; opencog-own-data tie pending).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

9923c73b06c33944d45958d2c22c28fbe38b8d99

by plankatron <fleeky@gmail.com> · 2026-07-05T16:50:07.000Z

Changed files (1) +9 -2

spec/convergence-with-linas.md +9 -2

…	…	@@ -120,8 +120,15 @@ Both prongs share one verification target per primitive.
120	120	on real data: GPU == cpu_mi == opencog `pair-fmi`.** (Key gotcha: compare to `pair-fmi`,
121	121	the pointwise MI, NOT `pair-mi`, the count-weighted one.) This is the cornerstone of
122	122	"reproducible of his results" — the MI primitive is fully data-verified.
123		2. Audit `gpu-sections.cl` — is its test tied to real `learn` pseudo-csets
124		(Section = word + ConnectorSeq, count on the Section)? Tie or fix.
	123	2. Audit `gpu-sections.cl` — PARTIAL 2026-07-05. Data model MATCHES opencog
	124	(section = `(word, disjunct)`, disjunct = sorted connector set, connector =
	125	`(partner, ±dir)`, count on section = `Section`/`ConnectorSeq`/`Connector`).
	126	`test-sections.c` verifies vs hand-computed ground truth (disjuncts + CPU hashes,
	127	dedup count=2.0, no cross-boundary) and builds+runs on the RTX 2070 → 6/6 PASS.
	128	OPEN: the disjunct is HASHED (FNV-1a) not symbolic, so the opencog data-tie is
	129	STRUCTURAL — run the same parse through opencog's pseudo-cset extraction and match
	130	`(word, connector-multiset, count)`. Needs the opencog MST pipeline (word-pair MI →
	131	MST → cset) — a sub-project, deferred. Sections are at MI's pre-1b bar.
125	132	3. Audit `gpu-connector-rewrite.cu` — does it implement the pre-sheaf
126	133	(type-class merge) mating semantics? Tie to opencog rewrite / matrix.
127	134	4. Selector stub — a `{Tₐ}` dispatch loop over the verified kernels: one fired