python scale and offset calculation for shrinking data to a specific integer

compute_scale_and_offset.py

import numpy as np
import math


def compute_scale_and_offset(min_value: (int, float, np.number), max_value: (int, float, np.number), dtype: np.dtype):
    """
    Computes the scale (slope) and offset for a dataset using a min value, max value, and the required np.dtype. 
    It leaves one value at the lower extreme to use for the nan fillvalue.
    These are the min values set asside for the fillvalue (up to 64 bits).
    int8:  -128
    int16: -32768
    int32: -2147483648
    int64: -9223372036854775808

    Unsigned integers are allowed and a value of 0 is set asside for the fillvalue.

    Parameters
    ----------
    min_value : int or float
        The min value of the dataset.
    max_value : int or float
        The max value of the dataset.
    dtype : np.dtype
        The data type that you want to shrink the data down to.

    Returns
    -------
    scale, offset as floats
    """
    bits = dtype.itemsize * 8
    data_range = max_value - min_value
    if bits < 64:
        target_range = 2**bits - 2
        target_min = -(2**(bits - 1) - 1)
        slope = data_range / target_range
    else:
        data_power = int(math.log10(data_range))
        target_range = 2**bits
        target_power = int(math.log10(target_range))
        target_min = -10**(target_power - 1)
        slope = 10**-(target_power - data_power)

    # Correction if the dtype is unsigned
    if dtype.kind == 'u':
        target_min = 1

    offset = min_value - (slope*target_min)

    return slope, offset