ndarray(numpy,python) assignment assigns array of 0s : why?

Question

So in this simple code the line

result[columnNumber] = column #this assignment fails for some reason!

fails, and specifically it simply assigns array of zeroes instead of what it is supposed to assign, and I have no idea why! So this is the full code:

    """Softmax."""

scores = [3.0, 1.0, 0.2]

import numpy as np

def softmax(x):
    """Compute softmax values for each sets of scores in x."""
    data=np.array(x)
    columnNumber=0
    result=data.copy()
    result=result.T

    for column in data.T:
        sumCurrentColumn=0
        try: #Since 'column' can potentially be just a double,and sum needs some iterable object
            sumCurrentColumn=sum(np.exp(column))
        except TypeError:
            sumCurrentColumn=np.exp(column)



        column=np.divide(np.exp(column),sumCurrentColumn)
        print(column)
        print('before assignment:'+str(result[columnNumber]))
        result[columnNumber] = column #this assignment fails for some reason!
        print('after assignment:'+str(result[columnNumber]))
        columnNumber+=1 

    result=result.T

    return result


scores = np.array([[1, 2, 3, 6],
                   [2, 4, 5, 6],
                   [3, 8, 7, 6]])

print(softmax(scores))

and this is its output:

   [ 0.09003057  0.24472847  0.66524096]
before assignment:[1 2 3]
after assignment:[0 0 0]
[ 0.00242826  0.01794253  0.97962921]
before assignment:[2 4 8]
after assignment:[0 0 0]
[ 0.01587624  0.11731043  0.86681333]
before assignment:[3 5 7]
after assignment:[0 0 0]
[ 0.33333333  0.33333333  0.33333333]
before assignment:[6 6 6]
after assignment:[0 0 0]
[[0 0 0 0]
 [0 0 0 0]
 [0 0 0 0]]

Warren Weckesser · Accepted Answer · 2016-08-11 10:27:33Z

2

In your example, the input scores is all integers, so the data type of the data array is integer. Therefore result is also an integer array. You can't assign a floating point value into an integer array--numpy arrays have homogeneous data types which can not be dynamically changed. The line result[columnNumber] = column is truncating the values in column to integers, and since they are all between 0 and 1, the assigned values are all 0.

Try changing the creation of result to:

result = data.astype(float)

(By default, the astype method creates a copy even if data already has the specified type.)

answered Aug 11, 2016 at 10:27

Warren Weckesser

116k20 gold badges207 silver badges224 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Yoni Keren Over a year ago

I feel a bit bad that I couldn't get it myself :) Thx man! By the way,since I don't know whether or not x is a multidimensional array or not, I thought I had to use that exception handling since sum() takes as an argument iterable, which a double or int is not. This feels not right though. Do you know of a more elegant way to do the same?

Julien · Accepted Answer · 2016-08-11 10:28:22Z

0

Your array result has type int so all your floats are automatically converted to int's, in this case 0. Use this result = data.astype(float).

answered Aug 11, 2016 at 10:28

Julien

15.3k6 gold badges33 silver badges58 bronze badges

Collectives™ on Stack Overflow

ndarray(numpy,python) assignment assigns array of 0s : why?

2 Answers 2

1 Comment

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Related