navlie.batch.problem.Problem¶

class navlie.batch.problem.Problem(solver: str = 'GN', max_iters: int = 100, step_tol: float = 1e-07, ftol: float | None = None, gradient_tol: float | None = None, tau: float = 1e-11, verbose: bool = True)¶

Bases: object

Main class for building nonlinear least squares problems.

is_converged(delta_cost, cost, dx, grad_norm) → bool¶

add_residual(residual: ~navlie.batch.residuals.Residual, loss: ~navlie.batch.losses.LossFunction = <navlie.batch.losses.L2Loss object>)¶

Adds a residual to the problem, along with a robust loss function to use. Default loss function is the standard L2Loss.

Parameters:

residual (Residual) – the error term to be added to the problem.
loss (LossFunction, optional) – robust loss to be used for this residual, by default L2Loss().

add_variable(key: Hashable, variable: State)¶: Adds a variable to the problem with a given key.

set_variables_constant(keys: List[Hashable])¶

Sets a variable to be held constant during optimization.

Parameters:: keys (List[Hashable]) – List of keys to be held constant

solve()¶: Solve the problem using either Gauss-Newton or Levenberg-Marquardt.

compute_error_jac_cost(variables: Dict[Hashable, State] | None = None) → Tuple[ndarray, ndarray, float]¶

Computes the full error vector, Jacobian, and cost of the problem.

Parameters:: variables (Dict[Hashable, State], optional) – Variables, by default None. If None, uses the variables stored in the optimizer.
Returns:: Error vector, Jacobian, and cost.
Return type:: Tuple[np.ndarray, np.ndarray, float]

get_covariance_block(key_1: Hashable, key_2: Hashable) → ndarray¶

Retrieve the covariance block corresponding to two variables.

Parameters:

key_1 (Hashable) – Key of first variable.
key_2 (Hashable) – Key of second variable.

Returns:

Covariance block corresponding to the two variables.

Return type:

np.ndarray

compute_covariance()¶: Compute covariance matrix after convergence of problem.

navlie.batch.problem.OptimizationSummary

navlie.batch.residuals