Check create_error_model

Check what does the BayesInference class attribute "use_error_model" does:

• It seems to call a "create_error_model" in the post_sampler parent class (the child classes have a similar version of it), which calls a "create_model_error" function in the surrogate class that no longer exists (it used to create a GP for the error, like PCE+GP)
• It needs a bias_inputs variable, which is always set to None.

added Inference Meta-modeling Restructuring labels

assigned to @mariafer.morales

marked this issue as related to #57 (closed)

changed title from create_error_model to Check create_error_model

changed milestone to %2.0.0

The built function is a replacement for gaussian properties of the surrogate. Try to replace it with MetaModel.is_gaussian.

Either always use the rmse if given, and not use the surrogate uq, or allow the user to choose between these two.

Set this as a flag, so this option does not get lost

When considering surrogate error, we can try to remove the loop over each sample to speed up the process using matrix multiplication

The create_error_model option was removed from BayesInf with !47 (merged)

closed