CODEWITHSUNDEEP
pythonmatlabnumpyscipy
Nils Werner
Nils Werner

Reputation: 36765

Identity element for `numpy.hstack`

How can I create an empty array wich I can then hstack with another array to fill with values?

For example in Matlab, I can do the following:

a = [];
b = [10 20];
a = [a b];

and would get

a = 

      10    20

I am looking for something similar in numpy. I tried

a = np.array([]);
b = np.array([10, 20]);
a = np.hstack((a, b)); # should be equal to `b`

But that gives

ValueError: all the input array dimensions except for the concatenation axis must match exactly

Upvotes: 0

Views: 5896

Answers (2)

Warren Weckesser
Warren Weckesser

Reputation: 114841

Using np.array([]) with hstack works for me.

In [11]: a = array([], dtype=int)

In [12]: b = array([10, 20])

In [13]: c = array([30, 40])

In [14]: a = hstack((a,b))

In [15]: a
Out[15]: array([10, 20])

In [16]: a = hstack((a,c))

In [17]: a
Out[17]: array([10, 20, 30, 40])

For vstack, the shape of the initial a needs some tweaking to make it have shape (0,2):

In [22]: a = array([], dtype=int).reshape(-1,2)

In [23]: a
Out[23]: array([], shape=(0, 2), dtype=int64)

In [24]: b
Out[24]: array([10, 20])

In [25]: c
Out[25]: array([30, 40])

In [26]: a = vstack((a,b))

In [27]: a
Out[27]: array([[10, 20]])

In [28]: a = vstack((a,c))

In [29]: a
Out[29]: 
array([[10, 20],
       [30, 40]])

Note that I've used dtype=int when creating the initial value of a. Without this, it uses the default dtype of float, and then when a is hstacked or vstacked with b, the result is upcast to float.

Upvotes: 2

Hooked
Hooked

Reputation: 88158

You can resize the dimension of a to match the shape of b and fill in the missing values. This will change the size of a and fill it with the default value, for np.floats this is 0.0

a = np.array([])
b = np.array([10, 20])
a.resize(b.shape)

np.hstack((a, b))

Upvotes: 2

Related Questions