symforce.opt.objectives.norm_barrier_objective module¶

class NormBarrierObjective[source]¶

Bases: TimestepObjective

class Params(norm_nominal, error_nominal, dist_zero_to_nominal)[source]¶

Bases: object

Parameters:

norm_nominal (float) –
error_nominal (float) –
dist_zero_to_nominal (float) –

norm_nominal¶

Value of the norm at which the error is equal to error_nominal.

Type:: float

error_nominal¶

Error returned when the norm is equal to norm_nominal.

Type:: float

dist_zero_to_nominal¶

The distance from norm_nominal to the region of zero error. Must be a positive number.

Type:: float

norm_nominal: float¶

error_nominal: float¶

dist_zero_to_nominal: float¶

class ExtraValues(unwhitened_residual)[source]¶

Bases: object

Parameters:: unwhitened_residual (Matrix11) –

unwhitened_residual¶

Error after applying the barrier function but before applying the cost scaling.

Type:: symforce.geo.matrix.Matrix11

unwhitened_residual: Matrix11¶

static residual_at_timestep(vector, params, epsilon, cost_scaling=1, power=1)[source]¶

Returns the residual block for the given timestep, where the residual is computed by applying a max barrier function to the norm of the given vector, and then optionally scaling the corresponding cost in the overall optimization problem by cost_scaling.

Parameters:

vector (Matrix) – Vector whose norm we wish to apply a barrier to.
params (Params) – Parameters defining the barrier function.
power (float) – Power of the barrier function. Defaults to 1, producing a linear barrier function in the residual, which corresponds to a quadratic cost in the overall optimization problem.
cost_scaling (float) – Optional scaling parameter. Corresponds to multiplying the cost in the overall optimization problem by a constant.
epsilon (float) –

Return type:

ResidualBlock