Coverage for src/cvxball/__init__.py: 100%

1"""Core package for minimum enclosing ball utilities and solvers. 

2 

3Exposes the version and both solvers, so the whole public surface is reachable as 

4``from cvxball import min_circle_active_set, min_circle_fgk``. The submodule paths 

5``cvxball.solver`` and ``cvxball.fischer_gaertner_kutz`` keep working, so this is 

6additive -- but the short form is the documented one, which leaves the module 

7layout free to change without breaking callers. 

8 

9The two solvers answer the same question and agree on the answer, arriving from 

10opposite sides: :func:`cvxball.min_circle_active_set` ascends the dual and holds 

11no enclosing ball until it terminates, while :func:`cvxball.min_circle_fgk` 

12deflates an enclosing ball and is feasible throughout. The first is the default -- 

13faster on every row of ``experiments/bench_seb.py``, and it returns the dual 

14weights as a certificate; :func:`cvxball.fischer_gaertner_kutz.ball_with_counts` 

15is the second one's fuller signature, reporting the support set and the pivot 

16counts alongside the ball. 

17 

18The dependencies are NumPy and SciPy, which the re-exports pull in on import. 

19That is the intended trade: they are the package's only reason to exist, so an 

20``import cvxball`` that did not pull them in would be deferring work every caller 

21is about to need. Nothing else is imported, because nothing else is needed -- the 

22Clarabel cone program and Welzl's recursion, which the two solvers are measured 

23against, live in ``experiments/`` and are references rather than solvers this 

24ships. 

25""" 

26 

27import importlib.metadata 

28 

29from cvxball.fischer_gaertner_kutz import Ball, ball_with_counts, min_circle_fgk 

30from cvxball.solver import min_circle_active_set 

31 

32__version__ = importlib.metadata.version("cvxball") 

33 

34__all__ = [ 

35 "Ball", 

36 "__version__", 

37 "ball_with_counts", 

38 "min_circle_active_set", 

39 "min_circle_fgk", 

40]