How to minmax normalize in the most efficient way, a XD-numpy array in "columns" of each 2D matrix of the array.
For example with a 3D-array :
a = np.array([[[ 0, 10],
[ 20, 30]],
[[ 40, 50],
[ 60, 70]],
[[ 80, 90],
[100, 110]]])
into the normalized array :
b = np.array([[[0., 0.],
[1., 1.]],
[[0., 0.],
[1., 1.]],
[[0., 0.],
[1., 1.]]])
With sklearn.preprocessing.minmax_scale + numpy.apply_along_axis single applying:
from sklearn.preprocessing import minmax_scale
a = np.array([[[0, 10], [20, 30]], [[40, 50], [60, 70]], [[80, 90], [100, 110]]])
a_scaled = np.apply_along_axis(minmax_scale, 1, a)
# a_scaled
[[[0. 0.]
[1. 1.]]
[[0. 0.]
[1. 1.]]
[[0. 0.]
[1. 1.]]]
a_min = a.min(axis=-2, keepdims=True)
a_max = a.max(axis=-2, keepdims=True)
out = (a - a_min) / (a_max - a_min)
out:
array([[[0., 0.],
[1., 1.]],
[[0., 0.],
[1., 1.]],
[[0., 0.],
[1., 1.]]])
Broadcasting and simple list comprehension
f= lambda ar:(ar==ar.max(axis=0)[None,:]).astype(int)
b = np.array([f(x) for x in a], dtype=float)
print(b)
This can also be done using numpy.apply_along_axis as follows:
ar = np.array([[[0, 10], [20, 30]], [[40, 50], [60, 70]], [[80, 90], [100, 110]]])
def f(a):
a = a.reshape(2,2)
return (a==a.max(axis=0)[None,:]).astype(int)
ar = ar.reshape(3,4)
b = np.apply_along_axis(f, 1, ar)
output
array([[[0., 0.],
[1., 1.]],
[[0., 0.],
[1., 1.]],
[[0., 0.],
[1., 1.]]])
