numpy type conversion during broadcast

Question

I'm using the numpy broadcast function to map a set of values over a set of coordinates. The values can be of heterogeneous type, including primitives. My problem is that the broadcast function is converting the primitive types under certain conditions. See for example this code:

In [11]: x = np.array([1])
    ...: y = np.array(['test', 10])
    ...: list(np.broadcast(x, y))

Out[11]: [(1, 'test'), (1, '10')]

The int 10 that was passed in the y value has been converted to a string '10' after the broadcast.

Is it possible to prevent this casting behaviour somehow?

It's not converted during the broadcast, it's converted at `y` initialization. Print `y` after your second line. — P. Camilleri, Jul 04 '17 at 12:06

score 2 · Accepted Answer · answered Jul 04 '17 at 12:11

2

This is because when you do

y = np.array(['test', 10])

the result is

array(['test', '10'], 
  dtype='<U4')

If you do y = np.array(['test', 10], dtype=object) the result will be as you wish.

answered Jul 04 '17 at 12:11

P. Camilleri

12,664
7
41
76

numpy type conversion during broadcast

1 Answers1