Describe the workflow you want to enable
Support Python typing, which is available in Python 3.5+, compatible with scikit-learn. This can enable tools to automatically get the estimator arg types (e.g., linters), for example. And it's a Python standard now.
Describe your proposed solution
For example, SVC could be like this:
def __init__(
self,
C: float = 1.0,
kernel: Union[Callable[[np.ndarray, np.ndarray], np.ndarray],
Literal["linear", "poly", "rbf", "sigmoid", "precomputed"]] = "rbf",
degree: int = 3,
gamma: Union[float, Literal["scale", "auto"]] = "scale",
coef0: float = 0.0,
shrinking: bool = True,
probability: bool = False,
tol: float = 1e-3,
cache_size: int = 200,
class_weight: Optional[Union[Mapping[int, float], Literal["balanced"]]] = None,
verbose: bool = False,
max_iter: int = -1,
decision_function_shape: Literal["ovo", "ovr"] = "ovr",
break_ties: bool = False,
random_state: Optional[Union[int, RandomState]] = None,
) -> None:Describe alternatives you've considered, if relevant
An alternative is to keep the types just in the docstrings. However, is hard for tools to parse the arg types.
Describe the workflow you want to enable
Support Python typing, which is available in Python 3.5+, compatible with scikit-learn. This can enable tools to automatically get the estimator arg types (e.g., linters), for example. And it's a Python standard now.
Describe your proposed solution
For example,
SVCcould be like this:Describe alternatives you've considered, if relevant
An alternative is to keep the types just in the docstrings. However, is hard for tools to parse the arg types.