src.blocks.ErrorModel module

This module is responsible for training and inference of error models.

It includes one class for the error models. See details in its own documentation.

Contact person: Stefan Riedmaier Creation date: 20.04.2020 Python version: 3.8

class src.blocks.ErrorModel.ErrorModel(config, domain)

Bases: object

This class is responsible for training and inference of error models.

It includes two main methods called “train_models” and “infer_models” that offer different meta-modeling techniques.

Most techniques are included from own external modules. In case of IPMs, two thin wrapper functions are required: train_ipm_model and infer_ipm_model.

See method details in their own documentations.

static infer_ipm_model(x, model)

This function performs inference using an Interval Predictor Model (IPM).

It uses the PyIPM package.

Parameters:

  • x (np.ndarray) – inputs

  • model – IPM model

Return tuple upper_bound, lower_bound:

upper and lower interval boundaries

infer_models(scenarios_da)

This function uses the trained error model to infer errors in the application domain for new scenarios.

Parameters:

scenarios_da – data array of scenarios

Returns:

static train_ipm_model(x, y, degree)

This function trains an Interval Predictor Model (IPM).

It uses the PyIPM package. The theory can be found, e.g., in [1] and [2].

Literature: [1] E Patelli, M Broggi, S Tolo, J Sadeghi, Cossan Software: A Multidisciplinary And Collaborative Software For Uncertainty Quantification, UNCECOMP 2017, At Rhodes Island, Greece, 2nd ECCOMAS Thematic Conference on Uncertainty Quantification in Computational Sciences and Engineering, June 2017. [2] L. G. Crespo, S. P. Kenny, D. P. Giesy, Y, R. B. Norman and S. R. Blattnig, „Application of Interval Predictor Models to Space Radiation Shielding,“ in 18th AIAA Non-Deterministic Approaches Conference, AIAA SciTech Forum, 2016.

Parameters:

  • x (np.ndarray) – inputs

  • y (np.ndarray) – outputs

  • degree (int) – degree of the polynomial

Returns:

trained interval predictor model

train_models(space_scenarios_da, metric_da)

This function trains an error model of validation metric results across the scenario space.

Parameters:

  • space_scenarios_da (xr.DataArray) – data array of scenarios

  • metric_da (xr.DataArray) – data array of validation metric results

Returns: