Getting value error while performing numpy.dot()

Question

I created two arrays using numpy:

import numpy as np    
a = np.array([[1, 5, 7], [6, 8, 9]])
b = np.array([[1, 8, 8], [5, 8, 0], [8, 9, 0]])
np.dot(a, b)

Now, while performing np.dot(a, b) I am getting the error :

ValueError: operands could not be broadcast together with shapes (2,3) (3,3) .

Normally, value error is raised if the last dimension of a is not the same size as the second-to-last dimension of b. What is wrong with my code?

For 2-D arrays it is equivalent to matrix multiplication. Please correct me if i am wrong. — Meghali Agrawal, Jan 26 '18 at 20:13
@MeghaliAgrawal You are correct. I wonder if there are version differences...Which one are you using? — Eugene Sh., Jan 26 '18 at 20:15
@EugeneSh. I am using python version 2.7. It worked with python version 3. Thanks for pointing it out. — Meghali Agrawal, Jan 26 '18 at 20:16
@MeghaliAgrawal please tell what do you get when you do `np.matmul(a, b)` ? — kmario23, Jan 26 '18 at 20:17
Works fine for me on Python 2.7.14 and NumPy 1.14.0. @MeghaliAgrawal: How did you install NumPy? — Mark Dickinson, Jan 26 '18 at 20:19
@MarkDickinson installed numpy using the command : pip install numpy — Meghali Agrawal, Jan 26 '18 at 20:22
I just wonder... Are you sure you are running this exact code? — Eugene Sh., Jan 26 '18 at 20:27
@MeghaliAgrawal how come you get different result for `matrix multiplication` ? ;) Please check your array inputs. Are you sure that you're running the code that you posted? — kmario23, Jan 26 '18 at 20:28
@kmario23 running the exact same code >>> import numpy as np >>> a = np.array([[1, 5, 7], [6, 8, 9]]) >>> b = np.array([[1, 8, 8], [5, 8, 0], [8, 9, 0]]) >>> np.dot(a, b) — Meghali Agrawal, Jan 26 '18 at 20:31
Something is fishy here... This is that rare case where a screenshot would be needed to convince... — Eugene Sh., Jan 26 '18 at 20:33
@EugeneSh. I am sorry. I don't know how to add images in comments. If you could help me with it, I will paste the screenshot. — Meghali Agrawal, Jan 26 '18 at 20:41
@MeghaliAgrawal You usually [edit] the question itself to make clarifications. — Eugene Sh., Jan 26 '18 at 20:42
@EugeneSh. I think I got what OP was doing wrong. It was due to using `*` operator for matrix multiplication. Now, I indeed get the error `ValueError: operands could not be broadcast together with shapes (2,3) (3,3) ` — kmario23, Jan 26 '18 at 20:57
@kmario23 Yes, I saw such a question here on SO ([here](https://stackoverflow.com/questions/24560298/python-numpy-valueerror-operands-could-not-be-broadcast-together-with-shapes)). But the OP is claiming otherwise....Anyway, since no clarification is coming, VTC as non-reproducible. — Eugene Sh., Jan 26 '18 at 20:58
`np.dot` does not use broadcasting, so shouldn't give this error. That makes me suspect that `np.dot` has been redefined by the users own code, or some module that has been imported. — hpaulj, Jan 27 '18 at 00:10
@MeghaliAgrawal [Please consider accepting the answer](https://stackoverflow.com/help/someone-answers) if it has helped else please let me know the issue. — kmario23, Apr 27 '19 at 14:16

kmario23 · Answer 1 · 2018-01-26T21:11:19.190

Your code works fine. Please note that when the inputs to np.dot() are matrices, np.dot() performs matrix multiplication

In [18]: a = np.array([[1, 5, 7], [6, 8, 9]])
    ...: b = np.array([[1, 8, 8], [5, 8, 0], [8, 9, 0]])
    ...: 

# @ is equivalent to `np.dot()` and `np.matmul()` in Python 3.5 and above
In [19]: a @ b
Out[19]: 
array([[ 82, 111,   8],
       [118, 193,  48]])


In [20]: (a @ b).shape
Out[20]: (2, 3)

# sanity check!
In [22]: a @ b == np.matmul(a, b)
Out[22]: 
array([[ True,  True,  True],
       [ True,  True,  True]], dtype=bool)

Note on @: It was introduced in Python 3.5 as a dedicated infix operator for matrix multiplication

This is because some confusion existed whether * operator does matrix multiplication or element-wise multiplication. So, to eliminate confusion, a dedicated operator @ was designated for matrix multiplication. So,

* performs element-wise multiplication
@ performs matrix multiplication (dot product)

The easiest way to perform a hassle-free dot product with two-dimensional NumPy arrays `a` and `b` is `(a * b).sum(axis=-1)` — brethvoice, Jul 21 '21 at 13:22

Getting value error while performing numpy.dot()

1 Answers1