Adding keyword arg to GLM for solver/regularizer

[BalzaniEdoardo]

Solver name should be added as keyword argument to GLM.

Regularizer can be passed as a string, or as a class. Default is UnRegularized, standard GLM.

Default for solver should be None and choose the default appropriately. Provide a nice table with the defaults for each regularizer. Raise an error if solver and regularizer are incompatible. Consider adding a flag that bypass the compatibility check (force_solver or something like that).

[BalzaniEdoardo]

Something like:

class BaseRegressor(Base, abc.ABC):

 def __init__(
      self,
      regularizer: str | Regularizer = "unregularized",
      solver: str = None

):

parse the regularizer and solver options

      # I think the parsing should just do string matching, so maybe we should add a class attribute to each regularizer 
      # that is the string equivalent 
      if not isinstance(regularizer, str):
           self._parse_options(regularizer=regularizer.name, solver=solver)

           # if no error returned, instantiate the solver because regularizer is already done
           self.solver = solver 
      else:
          self._parse_options(regularizer=regularize, solver=solver)

          # if no error raised, continue instantiating regularizer and solver 
          self.solver = solver # store the solver name so that it can be dynamically changed 

          self.regularizer = Regularizer(*kwargs) # instantiate instance

 def _parse_options(self, regularizer: str, solver: str):
      pass

 def _create_solver(*kwargs):
       # create solver would also need to parse the options just in case something changed 
       # create solver would get called by glm_fit() every time just in case changes have been made to the regularizer or the solver method
          return solver_run, solver_update, solver_init 

[clewis7]

BaseRegressor Changes

• add solver as str kwarg

  • default is None, use default_solver based on regularizer
  • parse to make sure that passed solver is valid for regularizer
  • allow dynamic changing of solver that gets updated when model.fit() is called

• add regularizer as kwarg that can be one of str or Regularizer object

  • default is "unregularized"
  • parse to make sure that passed regularizer is valid and can be instantiated
  • allow dynamic changing of solver that gets updated when model.fit() is called

TODO:

• [ ] fix GroupLasso so that when regularizer="group_lasso" is passed that there is a default mask
• [ ] add tests

[BalzaniEdoardo]

Solved in #183