Originally reported byLuca Geretti (Bitbucket: lgeretti, GitHub: lgeretti)
The implementation of uniform error reconditioning in geometry/enclosure.cc:1032 seems odd to me. I assume that it should work by collecting all the result indices for which a relatively large error is present in the model, then it modifies the models for those dimensions by creating a new parameter that encloses the error (which is then removed).
My doubts:
The error_domains variable is assigned twice with no particular reason.
The result dimensions chosen for moving the error into a new parameter do not match the dimensions for which the error is relatively large: we should apply the reconditioning to the large_error_indices[i] dimensions.
The second error>MAXIMUM_ERROR seems redundant to me.
Originally reported by Luca Geretti (Bitbucket: lgeretti, GitHub: lgeretti)
The implementation of uniform error reconditioning in
geometry/enclosure.cc:1032
seems odd to me. I assume that it should work by collecting all the result indices for which a relatively large error is present in the model, then it modifies the models for those dimensions by creating a new parameter that encloses the error (which is then removed).My doubts:
error_domains
variable is assigned twice with no particular reason.large_error_indices[i]
dimensions.error>MAXIMUM_ERROR
seems redundant to me.