Missing specialization for to_numpy for Series[float] and Series[int]

[cderemble]

For Series[float] and Series[int], the overloaded to_numpy functions are not expecting a np.dtype for the dtype arg. Maybe overloads missing ?

To Reproduce
Pyright will report an error for this line

pd.Series(np.empty(0, np.dtype("f8"))).to_numpy(dtype=np.dtype("f8"))

[loicdiridollou]

Thanks for the report @cderemble !
@Dr-Irv do you have any context on why DTypeLike has the np.dtype taken out of it?
Two options here:

• Add DTypeObj to DTypeLike which may be too much
• Add DTypeObj to the different overloads of the to_numpy methods and add another issue to look at it more in detail

[Dr-Irv]

Both you (@loicdiridollou ) and @cmp0xff have done most of the work on to_numpy. To handle the above, you'd want to do something similar to what is done here for strings:

    def to_numpy(
        self: Series[_str],
        dtype: NumpyStrDtypeArg,
        copy: bool = False,
        na_value: Scalar = ...,
        **kwargs: Any,
    ) -> np_1darray_str: ...

I don't think we want to do either of the 2 options that you have listed.

Having said that, I think the above declaration of dtype should be dtype: NumpyStrDtypeArg | | np.dtype[np.str_], so similar things would need to be done with the other declarations to include the correct numpy dtypes. What I'm not sure of is whether we want to (for example) change the definition of NumpyStrDtypeArg to include np.dtype[np.str_], or to only do it in the to_numpy overloads. If we expand the definition of the NumpyXXXDtypeArg types in _typing.pyi, it might make some of the other methods have too wide an overload. Not sure about that.

[cmp0xff]

I agree with @Dr-Irv for now, I think we need to fix to_numpy. Currently we don't have sufficient tests for the overloads.

import numpy as np
import pandas as pd


# everything runs
pd.Series([0.]).to_numpy(np.dtype("f8"))  # fails type checking
pd.Series([0]).to_numpy(np.dtypes.Int64DType)
pd.Series([0]).to_numpy(np.dtypes.Float64DType)
pd.Series([0]).to_numpy(np.dtypes.Float64DType())  # fails type checking
pd.Series([0]).to_numpy(np.dtypes.Int64DType())  # fails type checking
pd.Series([pd.Interval(1, 1)]).to_numpy(pd.IntervalDtype)
pd.Series([pd.Interval(1, 1)]).to_numpy(np.dtype("O")) # fails type checking