-
|
Hello ceres-team! The Is there a specific reason for only allowing Thanks for answers! |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 3 replies
-
|
Are you asking whether the The answer to the latter is that Ceres has been developed for solving real-valued problems, and the algorithms would often require some modification to correctly support complex valued types (e.g. (L)-BFGS has complex variants). Whilst that is doable, it would also have interface implications. Whilst it would be possible to template the parameter type and only explicitly instantiate it for a subset of types, ensuring correctness for all would be a significant overhead that we have not had a use-case for. |
Beta Was this translation helpful? Give feedback.
-
|
Yes I had complex types in mind. So allowing this would implicate quite some other changes. |
Beta Was this translation helpful? Give feedback.
-
|
Do you have a suggestions for c++ libraries for complex-valued nonlinear optimization problems? |
Beta Was this translation helpful? Give feedback.
-
|
Sorry but I am not aware of complex valued optimization libraries. |
Beta Was this translation helpful? Give feedback.
Are you asking whether the
parameterscould be another (real) floating point type (i.e.float) or whether it could be a complex type?The answer to the latter is that Ceres has been developed for solving real-valued problems, and the algorithms would often require some modification to correctly support complex valued types (e.g. (L)-BFGS has complex variants). Whilst that is doable, it would also have interface implications. Whilst it would be possible to template the parameter type and only explicitly instantiate it for a subset of types, ensuring correctness for all would be a significant overhead that we have not had a use-case for.