Slice 14 — Heatmap HM-1 (per-center viridis coloring) by 891458249 · Pull Request #13 · 891458249/RBF_MAX

891458249 · 2026-04-22T06:26:45Z

Summary

Phase 2B second slice. Adds the first viewport heatmap mode — HM-1: per-center weight visualization — so artists can see at a glance which trained centers carry the largest weights for the current rig.

mRBFNode (kDependNode)         ← +const weights() accessor
    │  message
    ▼
mRBFShape (kLocatorNode)       ← +heatmapMode enum (Off / CenterWeights / PredictionField)
    │  drawdb classification
    ▼
mRBFDrawOverride               ← per-center colored spheres via 11-stop viridis LUT, cached on weights buffer ptr

Deliverables

Phase 1: additive RBFInterpolator::weights() const getter + 1 D-group test (138 → 139)
adapter_core: new color_mapping module (Maya-free) — HeatmapMode enum, map_scalar_to_color, compute_center_colors — + 8 F-group tests
mRBFNode: additive weights() accessor (Phase 2A logic untouched)
mRBFShape: new aHeatmapMode enum attribute (3 fields; Slice 13 attrs untouched)
mRBFDrawOverride: per-center color path with cache key on weights buffer pointer; Slice 13 supportedDrawAPIs/boundingBox/isBounded/classification untouched
smoke_viewport: heatmapMode round-trip (Off ↔ CenterWeights)

F-stop #5 — polynomial → LUT pivot

The original spec called for a 4th-order polynomial fit of viridis. The candidate coefficients failed F1/F2/F3 by 0.038–0.486 absolute per channel against the spec's own reference values (the polynomial fits a different curve, not viridis). Per the spec's prescribed recovery path, pivoted to an 11-stop piecewise-linear LUT with anchors at v=0 / v=0.5 / v=1 matching F1/F2/F3 reference values exactly. Single rebuild → 172/172.

Validation

Check	Result
Adapter + Phase 1 (`build-adapter`, ctest)	172/172 passed
Maya 2022 plugin build (`C:/SDK/Maya2022/devkitBase`)	0 warn / 0 err
Maya 2025 plugin build (`C:/SDK/Maya2025/devkitBase`)	0 warn / 0 err
Maya 2022 × 4 smokes (hello/predict/train/viewport)	all PASS
Maya 2025 × 4 smokes (hello/predict/train/viewport)	all PASS
Phase 1 pure regression (`build`, ctest)	139/139 passed

Constraints respected

No Phase 1 kernel/solver/io_json behavioral changes (only additive const getter)
No Slice 10A/11/12 mRBFNode compute() / try_load() / train-cmd touches (only additive weights() accessor)
No Slice 13 mrbf_draw_override supportedDrawAPIs / boundingBox / isBounded / classification changes
No Slice 13 mrbf_shape pre-return MS::kSuccess line touches (new attribute appended at end of initialize())
No MRenderItem / custom shaders / playblast infra introduced
mRBFShape attribute set still strictly numeric + message + enum — no MFnTypedAttribute(kString), preserving the Slice 13 R-44 invariant

Tech-debt added

R-45: HeatmapMode::kPredictionField registered as enum field 2 but degrades to kOff in Slice 14; Slice 15 (HM-2) activates
T-19: per-frame min/max normalization makes color spread context-dependent
T-20: LUT precision 1e-2 at three anchors only; intermediate v values may drift ~0.05 vs matplotlib viridis

Test plan

Phase 2B Slice 14 (HM-1) groundwork. Mirror of Slice 13's centers() getter: const reference to FitResult::weights, noexcept, returns a 0x0 matrix when !is_fitted(). Exposes the (N x M) weight matrix to the Viewport 2.0 heatmap path so it can compute per-center L2 norms without re-parsing the saved JSON or re-running the solver. Phase 1 test count: 138 -> 139. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

… + F-group tests) Phase 2B Slice 14 (HM-1) — pure-C++14, Maya-free color-mapping module. Two surfaces: * map_scalar_to_color(Scalar v01) -> 4-tuple float RGBA. 11-stop piecewise-linear viridis LUT with anchors at v=0/0.5/1 matching matplotlib reference values exactly. Spec originally called for a quartic polynomial fit; pivoted to LUT after polynomial form failed F1/F2/F3 by ~0.4 absolute (see DEVLOG "F-stop #5"). * compute_center_colors(weights, out, n) — row L2 norm per center, per-frame min/max normalize, map through LUT. NaN/Inf weight in any row falls back to white for that single center (others use finite-only min/max). Degenerate "all weights equal" maps to the viridis start. 8 F-group tests cover both surfaces: F1/F2/F3 — viridis anchors at v=0/0.5/1 F4/F5 — clamping below 0 / above 1 F6 — alpha invariance across in/out-of-range inputs F7 — ascending norms map min->start max->end F8 — degenerate all-equal -> all colors equal & = viridis(0) Random seed kSeedS14 = 0xF5BFADu reserved for future randomised additions. Adapter test count: 25 -> 33 (E + F merged into the same target). Total adapter+Phase 1 ctest: 172/172 green. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…s / PredictionField) Phase 2B Slice 14 (HM-1) — adds the third (and currently last) attribute on mRBFShape: heatmapMode (enum, default 0=Off, channel-box visible) 0 = Off (Slice 13 default behaviour) 1 = Center Weights (Slice 14 HM-1) 2 = Prediction Field (reserved for Slice 15; falls back to Off) Field indices match the rbfmax::maya::HeatmapMode enum class so the DrawOverride can cast the short to the enum type directly. Slice 13 mRBFShape attributes (sourceNode, drawEnabled, sphereRadius) and its Path B classification + initialize() flow are untouched; the new block sits between addAttribute(aSphereRadius) and the trailing return MS::kSuccess as the spec requires. Strict invariant from Slice 13 Path A retrospective preserved: mRBFShape MUST NOT carry any MFnTypedAttribute(MFnData::kString) — the new attribute uses MFnEnumAttribute, which is unaffected by the empirical kLocatorNode-typed-string failure mode (DEVLOG R-44). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…aching Phase 2B Slice 14 (HM-1) — wires the heatmap end to end. Changes: - mRBFNode: additive const weights() returning the interpolator's N x M weight matrix (or a static empty 0x0 when !is_loaded()). Phase 2A compute() / try_load() / attribute set zero-touch; this is purely a read accessor for the DrawOverride's color path. - mRBFDrawOverride RbfDrawData: extended with center_colors vector, current heatmap_mode, and a 4-tuple cache key (weights buffer pointer + rows + cols + last cached mode). - prepareForDraw: reads heatmapMode plug; for kCenterWeights computes (or reuses cached) per-center colors via compute_center_colors; kPredictionField gracefully degrades to kOff. Cache key compares the underlying weights buffer pointer against the prior frame; pointer change (or shape change, or mode switch) triggers recompute. Slice 13 supportedDrawAPIs / boundingBox / isBounded / classification untouched. - addUIDrawables: switches to per-sphere setColor + sphere when center_colors is populated; otherwise keeps the Slice 13 single- color batch path. Per Maya 2022/2025 docs, calling setColor inside the begin/endDrawable block before each sphere() is the documented idiom for per-primitive coloring. - maya_node/CMakeLists.txt: src/color_mapping.cpp added to the plugin sources alongside the Slice 13 sources. - smoke_viewport.py: round-trips heatmapMode through Off (0) and CenterWeights (1) after the Slice 13 drawEnabled / sphereRadius asserts. Validation (8 smokes total, both Maya versions all PASS): Maya 2022 hellonode/predict/train/viewport: PASS Maya 2025 hellonode/predict/train/viewport: PASS Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

DEVLOG.md: ~140-line Slice 14 entry covering scope, deliverables, 12 design decisions, R-09 self-check (LUT precision, cache key soundness), validation outcomes (172 / 0 / 8 / 139), the F-stop #5 record (polynomial -> LUT pivot rationale + recovery), and tech- debt R-45/T-19/T-20. maya_node/README.md: new "Heatmap mode (Slice 14 - HM-1)" subsection inside the Slice 13 Viewport 2.0 section: enum table, workflow snippet, implementation notes (LUT precision contract, NaN fallback, per-frame normalization, cache key). Visual review screenshots (4: Maya 2022 kOff + kCenterWeights; Maya 2025 kOff + kCenterWeights) pending user-side GUI session and will be attached to PR description on capture. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…+ 2025) Captured during user-side GUI verification: - maya2022_Off.png — Maya 2022.5.1, heatmapMode=0 (4 white spheres) - maya2022_CenterWts.png — Maya 2022.5.1, heatmapMode=1 (3 deep-purple + 1 yellow) - maya2025_Off.png — Maya 2025.3, heatmapMode=0 (4 white spheres) - maya2025_CenterWts.png — Maya 2025.3, heatmapMode=1 (3 deep-purple + 1 yellow) Color distribution observation: tiny_rbf fixture's 4 weight-norm values cluster as {3 near min, 1 outlier}, which the per-frame min/max normalization maps to viridis(0)=purple (3 centers) and viridis(1)=yellow (1 center). No teal-band centers because the fixture lacks intermediate L2 norms. Production rigs (50+ centers) will exhibit the full purple->teal->yellow ramp. Cross-version invariant verified visually: both Maya 2022 and Maya 2025 produce identical color assignments at identical positions. This complements the deterministic adapter F-group tests by confirming no Maya-side color management drift between versions.

d891458249-rgb and others added 6 commits April 22, 2026 14:22

891458249 merged commit 9edffdd into main Apr 22, 2026
4 checks passed

891458249 deleted the slice-14-heatmap-hm1 branch April 22, 2026 08:01

891458249 mentioned this pull request Apr 23, 2026

Slice 16 — Phase 2B close-out + v1.2.0 #15

Merged

8 tasks

Navigation Menu

Slice 14 — Heatmap HM-1 (per-center viridis coloring) - #13

Slice 14 — Heatmap HM-1 (per-center viridis coloring)#13
891458249 merged 6 commits into
mainfrom
slice-14-heatmap-hm1

891458249 commented Apr 22, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

891458249 commented Apr 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Deliverables

F-stop #5 — polynomial → LUT pivot

Validation

Constraints respected

Tech-debt added

Test plan

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

891458249 commented Apr 22, 2026 •

edited

Loading