If the scale is different for each axis and it collapses when compared collectively, I would like to divide each axis by the maximum value or total value to normalize and compare.
You can do it with a for statement, but you can easily do it with ** keepdims = True **.
I added an axis after sum, but ** keepdims ** can do the same thing. I feel that there is something for this.
AAA = np.arange(24).reshape((2,3,4))
print("AAA.shape=", AAA.shape)
print("AAA=", AAA)
sAAA = AAA.sum(-1, keepdims=True)
# sAAA = AAA.sum(-1)[..., np.newaxis]Same as.
nAAA = AAA / sAAA
print("\nsAAA.shape=", sAAA.shape)
print("\nnAAA=", nAAA)
mAAA = AAA.max(1, keepdims=True)
# mAAA = AAA.max(1)[:, np.newaxis]Same as
nAAA = AAA / mAAA
print("\nmAAA.shape=", mAAA.shape)
print("\nnAAA=", nAAA)
mAAA = AAA.max((0,2), keepdims=True)
nAAA = AAA / mAAA
print("\nsAAA.shape=", sAAA.shape)
print("\nnAAA=", nAAA)
AAA.shape= (2, 3, 4)
AAA= [[[ 0 1 2 3]
[ 4 5 6 7]
[ 8 9 10 11]]
[[12 13 14 15]
[16 17 18 19]
[20 21 22 23]]]
sAAA.shape= (2, 3, 1)
nAAA= [[[0. 0.16666667 0.33333333 0.5 ]
[0.18181818 0.22727273 0.27272727 0.31818182]
[0.21052632 0.23684211 0.26315789 0.28947368]]
[[0.22222222 0.24074074 0.25925926 0.27777778]
[0.22857143 0.24285714 0.25714286 0.27142857]
[0.23255814 0.24418605 0.25581395 0.26744186]]]
mAAA.shape= (2, 1, 4)
nAAA= [[[0. 0.11111111 0.2 0.27272727]
[0.5 0.55555556 0.6 0.63636364]
[1. 1. 1. 1. ]]
[[0.6 0.61904762 0.63636364 0.65217391]
[0.8 0.80952381 0.81818182 0.82608696]
[1. 1. 1. 1. ]]]
sAAA.shape= (2, 3, 1)
nAAA= [[[0. 0.06666667 0.13333333 0.2 ]
[0.21052632 0.26315789 0.31578947 0.36842105]
[0.34782609 0.39130435 0.43478261 0.47826087]]
[[0.8 0.86666667 0.93333333 1. ]
[0.84210526 0.89473684 0.94736842 1. ]
[0.86956522 0.91304348 0.95652174 1. ]]]
Recommended Posts