Describe the feature or idea you want to propose
because we originally made ETS a standalone, user parameterised forecaster, we do all the validation internally. This is inefficient when searching, because it repeatedly does the same thing
Describe your proposed solution
Ultimately ETS can possibly be private and set through AutoETS. First I think switch
error_type: int | str = 1,
trend_type: int | str | None = 0,
seasonality_type: int | str | None = 0,
seasonal_period: int = 1,to just be ints then remove this code which is called on every fit
_validate_parameter(self.error_type, False)
_validate_parameter(self.seasonality_type, True)
_validate_parameter(self.trend_type, True)
# Convert to string parameters to ints for numba efficiency
def _get_int(x):
if x is None:
return 0
if x == ADDITIVE:
return 1
if x == MULTIPLICATIVE:
return 2
return x
self._error_type = _get_int(self.error_type)
self._seasonality_type = _get_int(self.seasonality_type)
self._trend_type = _get_int(self.trend_type)validate just once in AutoETS
Describe alternatives you've considered, if relevant
No response
Additional context
No response
Describe the feature or idea you want to propose
because we originally made ETS a standalone, user parameterised forecaster, we do all the validation internally. This is inefficient when searching, because it repeatedly does the same thing
Describe your proposed solution
Ultimately ETS can possibly be private and set through AutoETS. First I think switch
to just be ints then remove this code which is called on every fit
validate just once in AutoETS
Describe alternatives you've considered, if relevant
No response
Additional context
No response