What does [..., None] mean in Numpy?

Question

I am new to Numpy and deep learning. I encountered such a snippet when learning about layer normalization.

a=array([[[-0.66676328, -0.95822262,  1.2951657 ,  0.67924618],
        [-0.46616455, -0.39398589,  1.95926177,  2.36355916],
        [-0.39897415,  0.80353481, -1.46488175,  0.55339737]],
 
       [[-0.66223895, -0.16435625, -1.96494932, -1.07376919],
        [ 1.30338369, -0.19603094, -1.43136723, -1.0207508 ],
        [ 0.8452505 , -0.08878595, -0.5211611 ,  0.10511936]]])
u=np.mean(a, axis=(2,))
s = np.std(a, axis=(2,))
 
y = a-u[...,None]
y = y/s[...,None]
print(y)

########################Output###################################
array([[[-0.80954074, -1.12241971,  1.29657224,  0.63538821],
        [-1.0214588 , -0.96610083,  0.83874033,  1.14881929],
        [-0.30472338,  1.04125172, -1.49779981,  0.76127147]],
 
       [[ 0.46047519,  1.21440667, -1.51218696, -0.16269489],
        [ 1.56757537,  0.13400543, -1.04708279, -0.65449801],
        [ 1.53885365, -0.35203004, -1.2273397 ,  0.04051609]]])

What does [..., None] mean in Numpy? Thanks!

Does this answer your question? [What does the Ellipsis object do?](https://stackoverflow.com/questions/772124/what-does-the-ellipsis-object-do) — azro, Jan 08 '21 at 15:13
in short, it adds one more dimension. You can search `numpy broadcast` on SO or google to see as why we need to do so. — Quang Hoang, Jan 08 '21 at 15:13

score 1 · Accepted Answer · answered Jan 08 '21 at 15:14

1

None is an alias for the newaxis object. It creates an axis with length 1, essentially adding one more dimension. This can be useful for matrix multiplication etc.

answered Jan 08 '21 at 15:14

Edmund

332
1
9

"None is an alias for the newaxis object", more like the other way around! – Ivan Jan 08 '21 at 19:00

What does [..., None] mean in Numpy?

1 Answers1