OverviewFilesCommitsIssuesDocsReleasesPeers

atomspace-layer-discovery / commits

ref:
experiments/level2: RESULTS-LV — the scorer fix came after the first scoring, not before
e64df84 by plankatron · 2026-10-19
experiments/level2: Test LV-C results — to convergence the ported clusterer over-merges; C1-C4 FAIL
e60f958 by plankatron · 2026-10-19
experiments/level2: summarize-LVC.py (written before the LV-C run finished)
99794a1 by plankatron · 2026-10-18
experiments/level2: Test DICT results — dictionary lifts the MST by +0.003; candidates hold +0.10 it cannot find
4deb612 by plankatron · 2026-10-18
experiments/level2: PREREG Test LV-C — Linas's clusterer to convergence (all words, MI cutoff) (follows #18)
9eda613 by plankatron · 2026-10-18
experiments/level2: Test DICT (l2 --dict) — disjunct dictionary scoring over 31 candidate MSTs, Viterbi re-estimation
62f0642 by plankatron · 2026-10-18
experiments/level2: PREREG Test DICT — parse with the learned dictionary (disjunct-scored candidate MSTs) (follows #17, #18)
fd3faf3 by plankatron · 2026-10-18
experiments/level2: RESULTS-LV — restore the in-group-mi-cluster call lost to shell substitution
1c7176c by plankatron · 2026-10-18
experiments/level2: Test LV results — Linas's clusterer at 200/100 covers too little to be a layer
88245fa by plankatron · 2026-10-18
experiments: Test LV pipeline — l2 --sections/--labels, cluster.c --dump, lv.py token assignment + scores
1f8e6b6 by plankatron · 2026-10-18
experiments/level2: PREREG Test LV — Linas's clusterer (port, 200/100) as layer 1, intended vs as-run (follows #15, #17)
bd0d982 by plankatron · 2026-10-18
experiments/level2: re-parse loop results — classes lift the parse 0.570 -> 0.715 in one round; pair-MI MST caps at 0.735
1551a26 by plankatron · 2026-10-18
experiments/level2: re-parse loop (l2 --reparse): exchange over parse links, MST over classes, 5 rounds
fc40b6b by plankatron · 2026-10-18
experiments/level2: PREREG re-parse loop — classes from the parse, re-parse with classes, 5 rounds (follows #15, #16)
884512f by plankatron · 2026-10-18
experiments/level2: Test A-U results — undirected rule recovers 25 points; MST = forest; A1 still fails, by less
4d9d6cf by plankatron · 2026-10-18
experiments/level2: Test A-U — undirected FOREST/MST/MPG parses, uconn/ucomp rules, link F1 (l2 --undirected)
0c796d3 by plankatron · 2026-10-18
experiments/level2: PREREG Test A-U — undirected FOREST/MST/MPG parses, undirected unit rules (follows #15)
7725bc7 by plankatron · 2026-10-18
docs/reports: audit of the layer-discovery experiments against Linas's design
c5cd8e7 by plankatron · 2026-10-18
experiments/level2: correction notes — learned-parse comp numbers carry the arc-direction artifact
a1cd10d by plankatron · 2026-10-18
experiments: results — pooled-parse split (F1/F2 fail) and Test B (all fail); comp rule is direction-biased
0551093 by plankatron · 2026-10-18
experiments: pooled-parse sense split (jt --splitdiag2) and Test B unit choice (l2 --soft B rows), run + scorers
1bef2b4 by plankatron · 2026-10-18
experiments: PREREG sense split over pooled parses (F) and level2 Test B choosing units (G)
20f5836 by plankatron · 2026-10-18
experiments/level2: Test A-soft results — six parses beat one; count still the best recall ranking
7a26b70 by plankatron · 2026-10-18
experiments/jigsaw-transformer: v3 diagnosis — gold parse makes the split test exact; the learned parse breaks it
ae1fc3a by plankatron · 2026-10-18
experiments: v3 split diagnosis (jt --splitdiag) and level2 soft ranking (l2 --soft), run + scorers
7918707 by plankatron · 2026-10-18
experiments: PREREG v3 diagnosis (input swap 2x2) and level2 Test A-soft (rank over 6 parses)
c0a1768 by plankatron · 2026-10-18
experiments: results — level2 Test A-JT (B4-B6 pass, B1-B3 fail) and v3 sense split (all fail but audit)
1d19bb5 by plankatron · 2026-10-18
experiments: v3 sense split (jt --v3), JT layer-1 dump (jt --dump, l2 --jt), run + scorers
5bf8c7a by plankatron · 2026-10-18
experiments: PREREG level2 Test A-JT (JT layer 1) and jigsaw-transformer v3 (sense split by disjuncts)
1be28db by plankatron · 2026-10-18
experiments/jigsaw-transformer: v2 results — order-dependent per-occurrence effect, small next to class error
a0411ce by plankatron · 2026-10-18
experiments/jigsaw-transformer: v2 arms (--v2 word|dj), collapse rows, run + scorer
a2f3695 by plankatron · 2026-10-18
experiments/jigsaw-transformer: PREREG-v2 — word-neighbour and disjunct arms, collapse control
6c2c668 by plankatron · 2026-10-18
experiments/jigsaw-transformer: v1 results — PMI-emergent classes + soft parse work; sense claim not supported
3ffb8cb by plankatron · 2026-10-18
experiments/jigsaw-transformer: v1 learner (--v1): layer 0 from word-pair PMI, exchange over attended pairs
964b7e1 by plankatron · 2026-10-18
experiments/jigsaw-transformer: pre-registration v1 — classes emerge from word-level PMI
fc8ed86 by plankatron · 2026-10-18
experiments/jigsaw-transformer: v0 harness + smoke test — audit works, learner does not
946988d by plankatron · 2026-10-18
experiments/jigsaw-transformer: pre-registration v0 (before any code or run)
14bdcb0 by plankatron · 2026-10-18
docs/reports: prior work on a discovered level constraining the next search
c948ff2 by plankatron · 2026-10-18
experiments/fractal: run 2 results — BEST-G fails (significance favours the coarsest level)
7aedd3c by plankatron · 2026-10-18
experiments/fractal: BEST-G scorer (--rule bestg) for run 2; run 1 output unchanged
862a535 by plankatron · 2026-10-18
experiments/fractal: pre-registration 2 — combine levels by evidence strength (BEST-G)
4fb9add by plankatron · 2026-10-18
experiments/fractal: results — graded connectors as implemented mostly fail
5d99c5c by plankatron · 2026-10-18
experiments/fractal: harness (graded connectors: backoff parse + tiered level-2 ranking)
8dce47f by plankatron · 2026-10-18
experiments/fractal: pre-registration (before any code or run)
53f1644 by plankatron · 2026-10-18
experiments/ladder/port: fix rowp free after classes are added; round-2 diagnostics
ab92716 by plankatron · 2026-10-18
experiments/ladder/port: stage C parser statistical check PASS
19845da by plankatron · 2026-10-18
experiments/ladder/port: merge 1 compared with Linas; left-over pass is order-dependent
ae10fee by plankatron · 2026-10-18
experiments/ladder/port: stage E step 2b — merge loop (UNVERIFIED)
9bd6e22 by plankatron · 2026-10-18
experiments/ladder/port: stage E step 2a — seed, in-group, jaccard trim (as run)
8d12a21 by plankatron · 2026-10-18
experiments/ladder/port: stage E step 1 — covering matrix + similarities; 4th upstream break
f0955ec by plankatron · 2026-10-18