How to quantize a 4 dimensional array?

[PythonNPC]

Hello,

I am new to NumPy.

Lets say I have an array:

nparray = np.array(output2)
print(output2.shape)

Output:

(1, 26, 26, 128)

All the values are floating point numbers. Most of them are 0.xxxxxxx, few are between 1.xxxxxxx to 4.xxxxxxx and extreme rare ones are above 5.xxxxxxx. They can be negative or positive.


I want to quantize these values into 8bit integers from the range -127 to 128. So that the largest value in the array becomes a 128 and the smallest value becomes a -127. I know there might be resolution loss but I am okay with it.

How can I achieve this using NumPy (or some other technique)?

[deanhystad]

numpy.interp() can be used to re-interpolate an array of values from one range to another.

Output:
import numpy as np

floats = np.random.rand(10)
print(floats)
range_adjusted = np.interp(floats, (min(floats), max(floats)), (-128, 127))
ints = range_adjusted.astype("int8")
print(ints)

Output:
[0.23561182 0.47971123 0.04360216 0.56386097 0.78886284 0.9104152
 0.32166445 0.52518248 0.54179553 0.72678282]
[ -71    0 -128   25   91  127  -46   13   18   72]

The array has to be a single dimension, but you can reshape down to one, do the value adjustments, then reshape back to 4 dimensions.

[Gribouillis]

This function seems to work

>>> def func(a):
...     l, u = np.amin(a), np.amax(a)
...     return np.rint((a - l) * (255 / (u - l))) - 127
... 
>>> import numpy as np
>>> t = np.random.rand(5, 7)
>>> t
array([[0.48858387, 0.51885553, 0.50699018, 0.14738217, 0.105979  ,
        0.53540183, 0.86726475],
       [0.80267188, 0.57738387, 0.32951204, 0.71753945, 0.35037134,
        0.93836202, 0.18163814],
       [0.20919502, 0.27296382, 0.20278684, 0.82154427, 0.15883035,
        0.27340079, 0.82762161],
       [0.23281187, 0.54044188, 0.28653632, 0.79343013, 0.20178373,
        0.61876179, 0.47841846],
       [0.71916377, 0.81516669, 0.17768047, 0.11852501, 0.86190322,
        0.1823391 , 0.508792  ]])
>>> func(t)
array([[ -10.,   -1.,   -4., -114., -127.,    5.,  106.],
       [  86.,   17.,  -59.,   60.,  -52.,  128., -104.],
       [ -95.,  -76.,  -97.,   92., -111.,  -76.,   94.],
       [ -88.,    6.,  -72.,   84.,  -98.,   30.,  -13.],
       [  61.,   90., -105., -123.,  105., -104.,   -4.]])