Question

I want to create multiple matrices with numpy.zeros((N,M)). But I just realized that this not working as I though it would:

can someone explain me the outcome of the following code (1dim arrays for simplicity):

#!/usr/bin/python
import numpy as np

#sequential array creation 
X=np.zeros(1)
Y=np.zeros(1)
X[0],Y[0]=1.0,2.0
print X,Y

#multiple array creation
X,Y=[np.zeros(1)]*2
X[0],Y[0]=1.0,2.0
print X,Y

the result is

[ 1.] [ 2.]

[ 2.] [ 2.]

this means the second method to create the arrays does not work…

What is the prober way to create many ndarrays with identical dimensions in 1 line?

Answer 1

mylist * 2

is equivalent to

mylist + mylist #resulting list has 2 references to each element in mylist 
                #stored as:
                #[mylist[0],mylist[1],...,mylist[0],mylist[1],...]
                #   ^ ----------------------^
                #   reference the same object

So in your case, you’re making a numpy array, and then you’re putting it in a list. When you multiply that list, the resulting list has 2 references to the same array.

If you want to create multiple arrays and put them in a list, a list comprehension will do just fine:

lst_of_arrays = [ np.zeros(1) for _ in range(N) ]

Or if there are few enough of them to unpack, you can use a generator or a list-comprehension (below I opt for the generator):

X,Y = ( np.zeros(1) for _ in range(2) )
X,Y,Z = ( np.zeros(1) for _ in range(3) )
W,X,Y,Z = ( np.zeros(1) for _ in range(4) )
...

(and, in anticipation of an otherwise inevitable comment, in python2.x, you can use xrange instead of range to save the overhead of creating a list …)