Show posterior uncertainty in plot_model_cartoon as graded bands; declutter samples mode - #1130
Closed
AlexanderFengler wants to merge 9 commits into
Closed
Show posterior uncertainty in plot_model_cartoon as graded bands; declutter samples mode#1130AlexanderFengler wants to merge 9 commits into
AlexanderFengler wants to merge 9 commits into
Conversation
Both the predictive and model-cartoon plot families need the bandable step-curve geometry and the graded-band opacity ladder. Move _curve_xy out of predictive.py and extract the inline alpha linspace into _band_alphas, so the cartoon redesign consumes the same helpers instead of copying them. No behavior change: 44/44 fast plotting tests pass unchanged.
…#1124) plot_func_model gains the #1123 uncertainty vocabulary. RT histograms are now drawn from precomputed per-draw defective-density matrices (shared bin edges, _curve_xy geometry) instead of N overlaid ax.hist calls, enabling graded pointwise quantile bands, auto-alpha spaghetti, a real step/bins story, and a hist_height normalizer. Geometry uncertainty renders as boundary fan ribbons, a dashed drift cone (decision-time aligned, NaN-masked past absorption, truncated at <50% survival), graded ndt axvspans + rug replacing the per-draw axvline smear, and a start-point whisker. uncertainty=None preserves the legacy rendering for GUI-heritage callers; add_histograms_to_twin_axes is deleted; legend is assembled from labeled artists instead of hardcoded proxies.
Public signature adopts the #1123 vocabulary: uncertainty='band' default, hdi, alpha_mean/alpha_uncertainty, hist_height, legend, live bins/step/ colors/linestyles/linewidths (previously documented but silently dead). The deprecated plot_predictive_mean/plot_predictive_samples booleans map through a FutureWarning shim; passing both spellings errors. Geometry gets its own color_model (neutral black) so the figure encodes structure=neutral, predictions=predicted color, data=observed color. The drift cone truncates at the first absorption across draws — the >=50% survival rule showed selection-biased tails. Histogram baselines anchor to the same no-noise sim that draws the reference boundary, so histograms sit exactly on the drawn bound. Hard-coded proxy legends replaced by harvested, deduped, _legend_order-ranked legends on both the 1D and FacetGrid paths.
…#1124) plot_func_model_n adopts the same machinery: per-choice defective-density matrices (jointly integrating to 1 across choices) rendered as graded bands in the per-choice trajectory colors from the single shared baseline, plus a boundary ribbon and ndt spans/rug on the geometry. uncertainty=None keeps the legacy per-draw overlays. The rewrite removes the latent NameError when only observed data was passed (bottom now always defined), anchors the baseline to the reference no-noise boundary like the 2-choice renderer, and gives the legacy legend's dashed ndt proxy its missing label.
…ng (#1124) tests/unit/plotting/test_model_cartoon.py: 30 unmarked tests in the #1123 style — pure math pins (defective-density mass identities incl. -999 exclusion and non-uniform edges, per-choice masses, geometry matrices, band-alpha ladder, shim mappings) plus artist-census assertions per mode (collection/line counts, alpha values, zorders, legend labels, live bins/step/hist_height/styles, seeded reproducibility, legacy-census parity for uncertainty=None, and the n-choice data-only NameError regression). The slow structural suite converts to the uncertainty vocabulary, keeps one boolean case covering the shim end-to-end (FutureWarning + ValueError), and adds a facet-level PolyCollection check. tests/test_model_cartoon_colors.py is kept, deviating from the original plan: _add_trajectories and its colors support survived the redesign, so the test still covers live code.
docs/tutorials/cartoon_gallery.ipynb mirrors ppc_gallery.ipynb: fixture- driven (runs in seconds, no sampling), one option per section — default graded bands, samples, both, mean-only, custom hdi, hist_height, styling, facets, the deprecated-boolean shim, and the regression-model trial- provenance caveat. Executed end-to-end: 9 figures, zero errors. Wired into mkdocs nav + execute_ignore beside its sibling. Also fixes the plotting.ipynb cell that passed long-removed kwargs (plot_pp_mean/plot_pp_samples/alpha_pp, silently swallowed since their removal) and adds the changelog entry. Full re-execution of the cartoon sections in plotting.ipynb / scientific_workflow_hssm.ipynb is deferred to merge time to avoid double churn against the #1123 notebook regeneration.
end-of-file/whitespace auto-fixes in the two gallery notebooks, and the legend dedup rewritten as an explicit loop — mypy rejects the set.add()-in-boolean-context idiom. CI's gate (mypy src/hssm) is clean; the 14 remaining local-hook findings are pre-existing in tests/addm and tests/rl on the base branch.
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
Contributor
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
…#1124) Three fixes from first review of the gallery: 1. Warnings: attach_trialwise_params_to_df now sorts the (chain, draw, obs_n) MultiIndex once before its per-draw assignments — previously every assignment emitted a pandas lexsort PerformanceWarning (1236 per gallery run). The remaining numba object-mode notices trace to the fixture shipping no predictive groups (#1080); filtered in the gallery setup cell with a comment naming the cause. 2. Legend: the combined histogram + geometry legend is too tall to float inside the axes; it now sits outside on the right (center-left anchor at x=1.02). 3. Geometry coherence: all four no-noise geometry sims (reference and per-draw, both renderers) now use the trial-0 row as a single-row theta. Previously the reference simulated an unseeded RANDOM trial while the per-draw curves inherited the simulator's last-trial boundary scratch buffer — for hierarchical models the ribbons and the drawn mean bound sat at visibly different levels. One consistent trial also makes every metadata element (boundary/trajectory/ndt/z) describe the same trial and the reference deterministic. #1125 narrows to the reduction convention (trial-mean, obs=), seeded trajectories, and max_t. 4. NDT display: the graded axvspan envelope now renders in every uncertainty mode, replacing the bottom-edge rug in samples mode — an envelope hugging the dashed reference line is the natural reading; the rug read as a stray artifact. Gallery re-executed: zero warnings, 9 figures, no errors. 74/74 fast tests.
#1124) - _render_drift_band now draws fill_between quantile bands per HDI interval (widest first, _band_alphas ladder), matching the boundary ribbons and histogram bands; dashed linestyle is reserved for the non-decision-time reference line - band census test distinguishes boundary ribbons (zorder 1010) from drift bands (zorder 1012) and asserts no dashed non-vertical lines - gallery, changelog, and docstring wording updated; changelog note on trial provenance refreshed to the trial-0 convention
Member
Author
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #1124. Stacked on #1123 (reuses its helpers and
uncertainty=vocabulary); retarget tomainwhen #1123 merges.What changed
uncertainty="band" | "samples" | "both" | Noneonplot_model_cartoon, default"band"— same vocabulary asplot_predictive.Nonereproduces the old mean-only look (test-pinned).ax.histcalls (~0.98 accumulated opacity at defaults). Makes the previously deadbins/step/colors/linestyles/linewidthsparameters real; addshist_height.axvspans + rug replacing the per-draw axvline smear, starting-point whisker. Geometry stays neutral black (color_model); predictions/data keep the predictive palette.samplesmode: auto per-curve alphaclip(4/n_draws, .02, .25)instead of fixed 0.05.plot_predictive_mean/plot_predictive_samples→FutureWarning; both spellings together error.docs/tutorials/cartoon_gallery.ipynb(sibling ofppc_gallery.ipynb, executed, 9 figures). Changelog entry included.tests/unit/plotting/test_model_cartoon.py); slow suite converted to the new vocabulary + one shim-coverage case.Scope notes
plotting.ipynb/scientific_workflow_hssm.ipynb: stale removed-kwargs fixed in source; full re-execution deferred to merge time to avoid double churn against Show posterior uncertainty in plot_predictive by default; add samples and KDE displays #1123's notebook regeneration.Commands run
uv run pytest tests/unit/plotting/→ 74/74uv run mypy src/hssm(CI's gate) → clean;uv run ruff check .→ clean except a pre-existing unused import inppc_gallery.ipynbon the base branch🤖 Generated with Claude Code