numpy.ufunc.resolve_dtypes

method

ufunc.resolve_dtypes(dtypes, *, signature=None, casting=None, reduction=False)

Find the dtypes NumPy will use for the operation. Both input and output dtypes are returned and may differ from those provided.

Note

This function always applies NEP 50 rules since it is not provided any actual values. The Python types int, float, and complex thus behave weak and should be passed for “untyped” Python input.

Parameters:

dtypestuple of dtypes, None, or literal int, float, complex

The input dtypes for each operand. Output operands can be None, indicating that the dtype must be found.

signaturetuple of DTypes or None, optional

If given, enforces exact DType (classes) of the specific operand. The ufunc dtype argument is equivalent to passing a tuple with only output dtypes set.

casting{‘no’, ‘equiv’, ‘safe’, ‘same_kind’, ‘unsafe’}, optional

The casting mode when casting is necessary. This is identical to the ufunc call casting modes.

reductionboolean

If given, the resolution assumes a reduce operation is happening which slightly changes the promotion and type resolution rules. dtypes is usually something like (None, np.dtype("i2"), None) for reductions (first input is also the output).

Note

The default casting mode is “same_kind”, however, as of NumPy 1.24, NumPy uses “unsafe” for reductions.

Returns:

dtypestuple of dtypes

The dtypes which NumPy would use for the calculation. Note that dtypes may not match the passed in ones (casting is necessary).

See also

numpy.ufunc._resolve_dtypes_and_context

Similar function to this, but returns additional information which give access to the core C functionality of NumPy.

Examples

This API requires passing dtypes, define them for convenience:

>>> int32 = np.dtype("int32")
>>> float32 = np.dtype("float32")

The typical ufunc call does not pass an output dtype. np.add has two inputs and one output, so leave the output as None (not provided):

>>> np.add.resolve_dtypes((int32, float32, None))
(dtype('float64'), dtype('float64'), dtype('float64'))

The loop found uses “float64” for all operands (including the output), the first input would be cast.

resolve_dtypes supports “weak” handling for Python scalars by passing int, float, or complex:

>>> np.add.resolve_dtypes((float32, float, None))
(dtype('float32'), dtype('float32'), dtype('float32'))

Where the Python float behaves samilar to a Python value 0.0 in a ufunc call. (See NEP 50 for details.)