gllvm provides little access to the control argument of the numerical optimisation methods. It would sometimes be nice to configure these, for models that struggle to converge. In the case of #201 adjusting the "factr" argument of optim's L-BFGS-B method, or using the methods in https://cran.r-project.org/web/packages/minic/index.html) improves convergence. Todo:
- Provides better access to the control argument of the numerical optimizers used in the package
- Write a function that gets all the bits-and-pieces in the right place of the object after using an external optimizer
Both of these would be nice to have, but one of them should be sufficient.
Tagging @JenniNiku for awareness.
gllvm provides little access to the control argument of the numerical optimisation methods. It would sometimes be nice to configure these, for models that struggle to converge. In the case of #201 adjusting the "factr" argument of optim's L-BFGS-B method, or using the methods in https://cran.r-project.org/web/packages/minic/index.html) improves convergence. Todo:
Both of these would be nice to have, but one of them should be sufficient.
Tagging @JenniNiku for awareness.