Skip to content

PyMICE Vignette Parity — Post-release maintenance

Last updated: 2026-07-05

PyMICE 0.1.0 is publication-ready. See PUBLICATION.md for the release checklist.

Status snapshot: PARITY_STATUS.md records completed work. This file tracks non-blocking follow-ups only — cosmetic rendering, optional RNG backends, and documented sampler tolerances.

Tier A (report structure) is complete for V01–V08. Tier D (bit-for-bit RNG) is optional and documented as expected difference under default rng="numpy".

Recently completed (2026-07)

Item Vignette Notes
Draw-order goldens refresh V02–V04 regenerate_draw_order_goldens.py; 12 keys updated
Chain helpers wired V01–V06 vignette_rng.py; RNG audit steps match
print(imp) formatter V01 _mids_imputed_vars() excludes non-imputed age from visit sequence
Passive circular BMI V04 Visit order (hgt/wgt before bmi); constraint check exact
V05 step 16 golden V05 regenerate_v05_goldens.py; promoted to exact=True
V05 steps 21–26 labels V05 moment tolerance ~0.15 documented on multilevel diagnostics
Cox PH + pool + qbar V06 regenerate_v06_goldens.py; steps 11–13 exact=True
V06 δ-chains V06 run_v06_leiden_delta_chain() / run_v06_mammalsleep_delta_chain()
R ampute backend V07 PYMICE_R_AMPUTE; steps 4/9/10/11 exact with R
futuremice V08 parallel_mice(), process pool, ibind, parallelseed
V02 steps 7–8 V02 exact=True on mira/pool main blocks (goldens refreshed)
Label hygiene V05/V06 Stale draw-order partial_reason strings swept
Structural audit All audit_vignette 0 errors / 0 warnings
r_gaps closure All 35 methods registered; mids utilities
Optional backends — PYMICE_R_LMER, PYMICE_SKLEARN
V08 print fix V08 futuremice(print=False) no longer crashes

Active queue

P1 — Maintenance

  • [x] audit_rng_parity.py — 27/27 chain steps match (2026-07-05)
  • [x] maintain_parity.py — structural + RNG audit wrapper; refreshes parity_backlog.json

P2 — Optional algorithm / RNG

  • [ ] V01 steps 8–9: norm.predict OLS tolerance (optional R lm backend)
  • [ ] V07 steps 3.3, 5.5, 10.7: native ampute vs R 4.5.2 golden drift
  • [ ] V08 steps 5–7: pooled tables without R snapshot; parallelseed when unset

P3 — Cosmetic / rendering

  • [ ] Factor labels vs numeric codes in summary() / str() (V01, V03, V05)
  • [ ] matplotlib diagnostic parity notes (all vignettes) — shape match, not pixel-identical
  • [ ] V06 Kaplan–Meier / fluxplot rendering
  • [ ] V04 step 9 triple-passive iteration event log format vs R
  • [ ] V08 wall-clock benchmark figures (R-only; keep skipped)

P4 — Method fidelity (optional)

  • [x] r_gaps.py methods — all V02-listed imputation methods registered (2026-07-05)
  • [x] mids utilities — as_mids(), cbind_mids(), rbind_mids() in mids_construct.py
  • [x] Optional R lme4 backend for 2l.lmer / 2l.bin (PYMICE_R_LMER; NumPy fallback)
  • [x] lasso.* / lda sklearn auto-detect (PYMICE_SKLEARN)

Per-vignette remaining items

V01

  • Steps 8–9: norm.predict OLS tolerance (optional R lm backend)
  • Steps 11–12: PMM RNG under default rng="numpy" (exact with rng="r" + chain)

V02

  • Plots only (trace, stripplot)

V03

  • Factor label layout in summaries (cosmetic)

V04

  • Step 9 triple-passive: iteration event log format vs R
  • Diagnostic plots (matplotlib)

V05

  • Step 9.24: imputed norm summary atol=0.2 (documented)
  • Steps 21–26: diagnostic plots only; sampler moment tolerance ~0.15 (documented)

V06

  • Kaplan–Meier / fluxplot / δ diagnostic plots (cosmetic)

V07

  • Steps 3.3, 5.5, 10.7: native vs R 4.5.2 golden drift
  • Reference-only step 12 documents patterns / odds / run API sections

V08

  • Steps 5–7: pooled tables (no R snapshot); parallelseed without global seed

Verification

cd PyMICE
source ~/.venvs/brain-pymice/bin/activate
python devtools/maintain_parity.py
# or individually:
python devtools/audit_vignette_alignment.py
python devtools/audit_rng_parity.py
python devtools/regenerate_draw_order_goldens.py   # V02–V04
python devtools/regenerate_v05_goldens.py          # V05 step 16
python devtools/regenerate_v06_goldens.py          # V06 Cox/pool/qbar
pytest tests/vignettes/ tests/unit/test_vignette_blocks.py -q

References