CODEWITHSUNDEEP
pythonnumpy
FlorianFP
FlorianFP

Reputation: 31

Numpy sum keepdims error

Python throws an error when calling numpy sum function on a matrix. 

probs = exp_scores / np.sum(exp_scores, axis=1, keepdims=True)

The error

probs = exp_scores / np.sum(exp_scores, axis=1, keepdims=True)
TypeError: sum() got an unexpected keyword argument 'keepdims'

Context: Calculate the loss function for a softmax classifier. Numerator is the exponential of the score function for the correct class and denominator is the sum of all the exponentials for all possible classes.

Upvotes: 2

Views: 2081

Answers (2)

Mohamed Ali JAMAOUI
Mohamed Ali JAMAOUI

Reputation: 14699

The argument is valid in the latest version of numpy as explained here. Here is the full list of argument for numpy.sum:

numpy.sum(a, axis=None, dtype=None, out=None, keepdims=False)

This was added since version 1.7 as you can see in the source code here. So, you need to upgrade your numpy installation.

Upvotes: 2

MSeifert
MSeifert

Reputation: 152765

The keepdims argument was added in NumPy 1.7. At least the docstring of np.sum (1.6) hasn't listed it as one of the arguments:

numpy.sum(a, axis=None, dtype=None, out=None)

However the 1.7 docstring already listed it:

numpy.sum(a, axis=None, dtype=None, out=None, keepdims=False)

Given that NumPy 1.6 was released in 2012 you probably should update your NumPy package.

However you could also use np.expand_dims in case you can't (or don't want to) update NumPy:

np.expand_dims(np.sum(exp_scores, axis=1), axis=1)

Upvotes: 2

Related Questions