skgpuppy.InverseUncertaintyPropagation module

class skgpuppy.InverseUncertaintyPropagation.InverseUncertaintyPropagation(output_variance, gp, u, c, I, input_variances=None, upga_class=<class 'skgpuppy.UncertaintyPropagation.UncertaintyPropagationExact'>, coestimated=[])

Bases: object

Here, we optimize the cost function \(\sum_i = c_i n_i = \sum_i \frac{c_i}{v_i I_{ii}}\)

• \(c_i\): cost per sample for input i
• \(n_i\): number of samples for input i

Parameters:

• output_variance – desired maximum output variance
• gp – A gaussian process representing one output of one simulation
• u – Input vector where the uncertainty should be estimated
• upga_class – Class for uncertainty propagation with gaussian approximation
• c – cost vector for the input variances
• I – diagonal of the fisher Information matrix
• input_variances – None or list of known input variances (each unknown variance should be None in the list)
• coestimated – Variables, that are coestimated: list of lists of coestimated variables e.g. [[0,1],[2,3]] if x_0 and x_1 are parameters of the same distribution and x_2 and x_3 are from another distribution.

Warning

input_variances are ignored at the moment

get_best_solution()

Calculate the inverse uncertainty propagation.

Returns:Optimal variances, that lead to the desired output uncertainty with minimal sampling cost

class skgpuppy.InverseUncertaintyPropagation.InverseUncertaintyPropagationApprox(output_variance, gp, u, c, I, input_variances=None, coestimated=[])

Bases: skgpuppy.InverseUncertaintyPropagation.InverseUncertaintyPropagation

Parameters:

• output_variance – desired maximum output variance
• gp – A gaussian process representing one output of one simulation
• u – Input vector where the uncertainty should be estimated
• c – cost vector for the input variances
• I – diagonal of the fisher Information matrix
• input_variances – None or list of known input variances (each unknown variance should be None in the list)
• coestimated – Variables, that are coestimated list of lists of coestimated variables

:note input_variances are ignored at the moment

get_best_solution()

class skgpuppy.InverseUncertaintyPropagation.InverseUncertaintyPropagationNumerical(output_variance, gp, u, c, I, input_variances=None, upga_class=<class 'skgpuppy.UncertaintyPropagation.UncertaintyPropagationExact'>, coestimated=[])

Bases: skgpuppy.InverseUncertaintyPropagation.InverseUncertaintyPropagation

Parameters:

• output_variance – desired maximum output variance
• gp – A gaussian process representing one output of one simulation
• u – Input vector where the uncertainty should be estimated
• upga_class – Class for uncertainty propagation with gaussian approximation
• c – cost vector for the input variances
• I – diagonal of the fisher Information matrix
• input_variances – None or list of known input variances (each unknown variance should be None in the list)
• coestimated – Variables, that are coestimated: list of lists of coestimated variables e.g. [[0,1],[2,3]] if x_0 and x_1 are parameters of the same distribution and x_2 and x_3 are from another distribution.

Warning

input_variances are ignored at the moment

get_best_solution(startvalue=None)

See baseclass method

Parameters:startvalue – Supply a startvalue for the optimization