Hey guys,
If I set a block constant using SetParameterBlockConstant, are the derivatives still computed with respect to this parameter block or is the optimization really treated as if these parameters are constants?
My motivation is the following: I want to optimze against a set of parameter blocks using Levenberg Marquardt. To improve convergence I set everything constant except for the first block and then optimize, then I set the first to blocks to be variable , optimize and so on. If the derivatives would be computed w.r.t. all parameters I would expect convergence to be worse than using a loss function, where the fixed parameter blocks really are constants.
Greetings
Janos