sciencehub-team/constellarationPublic

History / version R0DWA469 · saved 1 day ago, the current version

Cite this version

BibTeXAPACFF

@software{constellaration_2026,
  author = {ScienceHub editor},
  title = {ConStellaration: evaluate},
  version = {R0DWA469},
  year = {2026},
  publisher = {ScienceHub},
  url = {https://sciencehub.sh/sciencehub-team/constellaration/history/ver_01M61HRZRN56BCM56JR0DWA469}
}

editor, S. (2026). ConStellaration: evaluate (Version R0DWA469) [Computer software]. ScienceHub. https://sciencehub.sh/sciencehub-team/constellaration/history/ver_01M61HRZRN56BCM56JR0DWA469

cff-version: 1.2.0
message: "If you use this version, cite it as below."
title: "ConStellaration: evaluate"
type: software
authors:
  - family-names: "editor"
    given-names: "ScienceHub"
version: "R0DWA469"
date-released: "2026-10-28"
url: "https://sciencehub.sh/sciencehub-team/constellaration/history/ver_01M61HRZRN56BCM56JR0DWA469"

This address names exactly this version. The project page cites all versions.

evaluate.py

Download Open and run

3 cells changed against the version before: +7 −6 lines.

25 unchanged cells

ChangedCell+1−1

1 unchanged line
 for _key, _spec in PROBLEMS.items():
     for _name, _relation, _bound in _spec["constraints"]:
         _rows.append({"problem": _spec["title"], "objective": f"{_spec['objective'][1]} {_spec['objective'][0]}", "constraint": f"{_name} {_relation} {_bound:g}"})
-Removed: mo.ui.table(pd.DataFrame(_rows), selection=None, page_size=15, label="Constraints of constellaration 0.2.6")
+Added: mo.vstack([mo.md("Constraints of constellaration 0.2.6"), mo.Html(pd.DataFrame(_rows).to_html(index=False, border=0))])

ChangedCell+4−3

1 unchanged line
 ## Run the evaluator
  
 One evaluation is a full equilibrium solve at the leaderboard's
-Removed: resolution. On one x86-64 core it takes about a minute for the example
-Removed: boundary and up to about four minutes for the largest boundaries on the
-Removed: leaderboard. The two QI problems also compute a Boozer
+Added: resolution. On one x86-64 core of our CI runners it takes about a minute
+Added: for the example boundary and up to about four minutes for the largest
+Added: boundaries on the leaderboard. On a ScienceHub kernel the example
+Added: boundary took 189 s. The two QI problems also compute a Boozer
 transform and the QI residual, which need about 3 GiB of memory, so on a
 ScienceHub kernel (1.5 GiB) this notebook refuses them. The geometrical
 problem needs 1.0 GiB for boundaries of the dataset's size. The largest
10 unchanged lines

2 unchanged cells

ChangedCell+2−2

9 unchanged lines
         }
         for p in result["points"]
     ])
-Removed:     _view = mo.vstack([mo.md(f"Hypervolume score: **{result['score']:.3f}**"), mo.ui.table(_points, selection=None)])
+Added:     _view = mo.vstack([mo.md(f"Hypervolume score: **{result['score']:.3f}**"), mo.Html(_points.to_html(index=False, border=0, na_rep="", float_format=lambda _x: f"{_x:.4g}"))])
 elif "error" in result:
     _view = mo.md(f"The evaluation failed after {result['seconds']:.0f} s and scores 0: `{result['error']}`")
 else:
3 unchanged lines
             f"({'feasible' if result['feasible'] else 'not feasible'}), score **{result['score']:.4f}**. "
             f"The evaluation took {result['seconds']:.0f} s."
         ),
-Removed:         mo.ui.table(pd.DataFrame(result["constraints"]), selection=None),
+Added:         mo.Html(pd.DataFrame(result["constraints"]).to_html(index=False, border=0, float_format=lambda _x: f"{_x:.4g}")),
     ])
 _view

3 unchanged cells