CODEWITHSUNDEEP
pythonnumpymatrix
TabeaY.
TabeaY.

Reputation: 33

Why does numpy.copy of a numpy.matrix not act like the original matrix? Multiplication with the transpose of that copy does not work

I need to work with copys of matrices inside functions. But the copy of a (n x 1) matrix (vector) doesn't behave like it should.

Here I made an example:

Transpose of x multiplied with y gives me a normal vector-multiplication with an outcome of a (1x1)-matrix.

The copys a and b of x and y won't do that. They give back an array with dimension (n x n). What am I doing wrong here? And how could I avoid that?

    >>>import numpy as np

    >>>x=np.matrix('1;2;3')
    >>>y=np.matrix('1;1;-1')

    >>>x.T*y
    matrix([[0]])

    >>>a=np.copy(x)
    >>>b=np.copy(y)

    >>>a.T*b
    array([[ 1,  2,  3],
           [ 1,  2,  3],
           [-1, -2, -3]])

Upvotes: 2

Views: 5747

Answers (2)

donkopotamus
donkopotamus

Reputation: 23196

If you wish to copy a matrix, then instead of using numpy.copy, use the copy method on matrix. 

>>> x = np.matrix('1;3;3')
>>> x.copy()
matrix([[1],
        [3],
        [3]])

Another alternative is to use numpy.array(x, copy=True, subok=True).

Note that numpy.copy is simply an alias for numpy.array(x, copy=True), and this causes downcasting of the input.

Upvotes: -1

hpaulj
hpaulj

Reputation: 231425

Your original arrays are of subclass matrix. The copy is the base array class. Use x.copy(), the copy method specific to the matrix class to make another matrix. Then the matrix multiplication operations will work as before.

In [52]: x=np.matrix('1;3;3')
In [53]: x
Out[53]: 
matrix([[1],
        [3],
        [3]])
In [54]: np.copy(x)
Out[54]: 
array([[1],
       [3],
       [3]])
In [55]: x.copy()
Out[55]: 
matrix([[1],
        [3],
        [3]])

The solution proposed in the other answer is to replace the matrix multiplications with the equivalent ones for np.array (np.dot).

Upvotes: 3

Related Questions