CODEWITHSUNDEEP
pythonnumpymatrixarray-broadcasting
ShanZhengYang
ShanZhengYang

Reputation: 17621

np.dot() with Python broadcasting

I have two numpy arrays, one shaped (3000,) and the other is an array of twenty 3000 by 3000 matrices, i.e. shape (20, 3000, 3000)

first.shape = (3000,)
second.shape = (20, 3000, 3000)

I being doing a numpy dot product. 

import numpy as np
dotprod1 = np.dot( second, first)

this works, and the output dotprod1 is an array shaped (20, 3000).

But what if I wish to take the dot product again? 

dotprod2 = np.dot( first, dotprod1)

This gives an error. 

ValueError: shapes (3000,) and (20,3000) not aligned: 3000 (dim 0) != 20 (dim 0)

I would like to have an output of 20 values. How does one use broadcasting to do this?

Upvotes: 3

Views: 8507

Answers (1)

zhangxaochen
zhangxaochen

Reputation: 33997

dotprod2 = np.dot( first, dotprod1) fails because first is of shape (3000, ) and dotprod1 is of shape (20, 3000), swap them and the error will go (if that's your intention):

dotprod2 = np.dot(dotprod1, first)

besides, you can also use np.ndarray.dot to make the semantics clear:

dotprod2 = dotprod1.dot(first)

Upvotes: 3

Related Questions