Source code for honeybee_radiance_postprocess.data_type

"""Functions for NumPy data type (dtype)."""
from typing import Tuple
import numpy as np




def smallest_integer_dtype(array: np.ndarray) -> np.signedinteger:
    """Return the smallest possible integer dtype.

    Args:
        array: NumPy array.
    
    Returns:
        A NumPy integer dtype.
    """
    if np.all(array >= np.iinfo(np.int8).min) and \
        np.all(array <= np.iinfo(np.int8).max):
        return np.int8
    elif np.all(array >= np.iinfo(np.int16).min) and \
        np.all(array <= np.iinfo(np.int16).max):
        return np.int16
    elif np.all(array >= np.iinfo(np.int32).min) and \
        np.all(array <= np.iinfo(np.int32).max):
        return np.int32
    elif np.all(array >= np.iinfo(np.int64).min) and \
        np.all(array <= np.iinfo(np.int64).max):
        return np.int64





def smallest_float_dtype(array: np.ndarray, rtol: float = 1e-5,
                         atol: float = 1e-5) -> np.floating:
    """Return the smallest possible float dtype.

    The allclose function is used to check if a certain floating-point precision
    can be used without losing accuracy.

    Args:
        array: NumPy array.
        rtol: The relative tolerance parameter for `np.allclose`. The default
            is 1e-5.
        atol: The absolute tolerance parameter for `np.allclose`. The default
            is 1e-5.
    
    Returns:
        A NumPy floating dtype.
    """
    if np.all((array >= np.finfo(np.float16).min) & \
                (array <= np.finfo(np.float16).max)):
        if np.allclose(array, array.astype(np.float16), rtol=rtol, atol=atol):
            return np.float16
    if np.all((array >= np.finfo(np.float32).min) & \
                (array <= np.finfo(np.float32).max)):
        if np.allclose(array, array.astype(np.float32), rtol=rtol, atol=atol):
            return np.float32
    if np.all((array >= np.finfo(np.float64).min) & \
                (array <= np.finfo(np.float64).max)):
        if np.allclose(array, array.astype(np.float64), rtol=rtol, atol=atol):
            return np.float64





def smallest_dtype(array: np.ndarray, rtol: float = 1e-5, atol: float = 1e-5
                   ) -> Tuple[np.signedinteger, np.floating]:
    """Return the smallest possible dtype.

    Args:
        array: NumPy array.
        rtol: The relative tolerance parameter for `np.allclose`. The default
            is 1e-5. This is also used if the dtype of the array is np.floating.
        atol: The absolute tolerance parameter for `np.allclose`. The default
            is 1e-5. This is also used if the dtype of the array is np.floating.
    
    Returns:
        A NumPy dtype.
    """
    if np.issubdtype(array.dtype, np.integer):
        return smallest_integer_dtype(array)
    elif np.issubdtype(array.dtype, np.floating):
        return smallest_float_dtype(array, rtol=rtol, atol=atol)
    else:
        raise TypeError(f'Expected integer or floating dtype. Got {array.dtype}')





def set_smallest_dtype(array: np.ndarray, rtol: float = 1e-5,
                       atol: float = 1e-5) -> np.ndarray:
    """Return a NumPy array with the smallest possible dtype.
    
    Args:
        array: NumPy array.
        rtol: The relative tolerance parameter for `np.allclose`. The default
            is 1e-5. This is also used if the dtype of the array is np.floating.
        atol: The absolute tolerance parameter for `np.allclose`. The default
            is 1e-5. This is also used if the dtype of the array is np.floating.
    
    Returns:
        A new NumPy array with a smaller dtype.
    """
    dtype = smallest_dtype(array, rtol=rtol, atol=atol)
    return array.astype(dtype)