LinearSolverState¶

class probnum.linalg.solvers.LinearSolverState(problem, prior, rng=None)[source]¶

Bases: object

State of a probabilistic linear solver.

The solver state separates the state of a probabilistic linear solver from the algorithm itself, making the solver stateless. The state contains the problem to be solved, the current belief over the quantities of interest and any miscellaneous quantities computed during an iteration of a probabilistic linear solver. The solver state is passed between the different components of the solver and may be used internally to cache quantities which are used more than once.

Parameters

problem (LinearSystem) – Linear system to be solved.
prior (LinearSystemBelief) – Prior belief over the quantities of interest of the linear system.
rng (Optional[Generator]) – Random number generator.

Attributes Summary

`action`	Action of the solver for the current step.
`actions`	Actions taken by the solver.
`observation`	Observation of the solver for the current step.
`observations`	Observations of the problem by the solver.
`residual`	Cached residual \(Ax_i-b\) for the current solution estimate \(x_i\).
`residuals`	Residuals \(\{Ax_i - b\}_i\).

Methods Summary

next_step()

Advance the solver state to the next solver step.

Attributes Documentation

action¶

Action of the solver for the current step.

Is None at the beginning of a step and will be set by the policy.

Return type: Optional[ndarray]

actions¶

Actions taken by the solver.

Return type: Tuple[ndarray, …]

observation¶

Observation of the solver for the current step.

Is None at the beginning of a step, will be set by the observation model for a given action.

Return type: Optional[Any]

observations¶

Observations of the problem by the solver.

Return type: Tuple[Any, …]

residual¶

Cached residual \(Ax_i-b\) for the current solution estimate \(x_i\).

Return type: ndarray

residuals¶

Residuals \(\{Ax_i - b\}_i\).

Return type: Tuple[ndarray, …]

Methods Documentation

next_step()[source]¶

Advance the solver state to the next solver step.

Called after a completed step / iteration of the probabilistic linear solver.

Return type: None