conatus.expert.Ensemble

class conatus.expert.Ensemble(solver, size, *, rhs)

Bases: object

B members of one discretized problem advanced together (see module docstring).

Parameters:

• size (int)

• rhs (str)

classmethod from_solvers(solvers, *, rhs, member_of=None)

Parameter ensemble: member b uses solvers[member_of[b]] (default: one member per solver). The solvers must share the discretization, scheme, dt, start-up and backend.

Parameters:

• solvers (Sequence)

• rhs (str)

• member_of (Sequence[int] | None)

Return type:

Ensemble

set_states(X, t=0.0)

Initial states (B, n) (global vectors, e.g. solver.X of set_state calls); resets the multistep history.

Parameters:

t (float)

Return type:

None

F(X)

rk_stages(x0, dt, tableau)

Stage states of one IMEX RK step for the batch: same arithmetic, same order as IVPSolver.rk_stages (stage j’s F and L X computed only if a later stage uses them).

Parameters:

dt (float)

Return type:

list

step()

One step for every member (mirror of IVPSolver.step).

Return type:

None

run(steps, callback=None, every=None)

Advance steps steps; callback(self) after every step whose iteration is a multiple of every (both explicit, as in IVPSolver.run).

Parameters:

• steps (int)

• callback (Callable[[Ensemble], None] | None)

• every (int | None)

Return type:

None

states()

Host copy of the member states (B, n): a copy on every backend (finding 68: on NumPy the live state array was returned, so writing to it changed the ensemble).

Return type:

ndarray

explain()

Return type:

str