CODEWITHSUNDEEP
pythonnumpyvectorization
guroosh
guroosh

Reputation: 652

Numpy vectorize() is flattening the whole array

My input is a numpy array of tuples 

values = np.array([(4, 5, 2, 18), (4, 7, 3, 8)])

and my function is as follows:

def outerFunc(values):
    print(values)
    def innerFunc(values):
        print(values)
        mean = np.mean(values)
        result = 0
        for i in range(len(values)):
            result += math.pow(values[i] - mean, 2)
        return result

    if isinstance(values, np.ndarray):
        return np.vectorize(innerFunc)(values)
    else:
        return innerFunc(values)

Although I want to vectorize over 1 dimension, i.e., one tuple is executed inside the innerFunc, but my output is as follows:

[[ 4  5  2 18]
 [ 4  7  3  8]]
4
...

Which means the vectorize function is vectorizing over 2 dimensions, and I am getting the following error:

for i in range(len(values)):
TypeError: object of type 'numpy.int64' has no len()

What changes to make so that the output is:

[[ 4  5  2 18]
 [ 4  7  3  8]]
[4  5  2 18]
...

something like this

Thank you.

EDIT

It is working as accepted when the tuples are different length, can anyone explain this,

e.g., my input is 

np.array([(4, 5, 2, 18), (4, 7, 3,)])

and the function prints

[(4, 5, 2, 18) (4, 7, 3)]
(4, 5, 2, 18)
(4, 7, 3)

and the returned value is

[158.75         8.66666667]

So, only when all the tuples are the same length, the function treats them as numbers.

Thank you.

Upvotes: 2

Views: 1930

Answers (1)

hpaulj
hpaulj

Reputation: 231500

In [1]: values = np.array([(4, 5, 2, 18), (4, 7, 3, 8)])                        
In [2]: values                                                                  
Out[2]: 
array([[ 4,  5,  2, 18],
       [ 4,  7,  3,  8]])
In [3]: values.shape                                                            
Out[3]: (2, 4)
In [4]: x=np.array([(4, 5, 2, 18), (4, 7, 3,)])                                 
In [5]: x                                                                       
Out[5]: array([(4, 5, 2, 18), (4, 7, 3)], dtype=object)
In [6]: x.shape                                                                 
Out[6]: (2,)

values is a 2d numeric array. np.vectorize passes each of the 8 elements, one at a time, to your inner function. It does not iterate by rows.

x is a 1d array with 2 elements (tuples). vectorize will pass each of those tuples to your inner.

Don't use vectorize when a simple iteration would work - it's slower and harder to use right.

And look at your arrays after you create them, making sure you understand the shape and dtype. Don't make assumptions.

Upvotes: 3

Related Questions