casadi.reshape not compatible with numpy.reshape

[harzer...@gmail.com]

Hi,

during my work I ran into some issue (that are now solved) that were due to the different behavior of casadi.reshape() and numpy.reshape(). As in the example, I tried to reshape a matrix into a vector It looks like casadi shapes columns first and numpy rows first.

Sample Code

import casadi as ca
import numpy as np

A = ca.MX.sym('A',(2,2))
casadiReshape = ca.Function('caRes',[A],[A.reshape((4,1))])
print(f"Matrix: {np.array([[1, 2],[3, 4]])}")
print(f"Casadi: {casadiReshape(np.array([[1, 2],[3, 4]]))}")
print(f"Numpy: {np.array([[1, 2],[3, 4]]).reshape((4,1))}")

Prints:

Matrix: [[1 2]

              [3 4]]

Casadi:  [1, 3, 2, 4]

Numpy: [[1]

 [2]

 [3]

 [4]]

Its not an issue any more, I just wanted to ask if this is intended?

Cheers Jakob

[Kirill Klimov]

Hi! I'm not a CasADi dev, but as far as i know it's intended. NumPy uses row major data storage but CasADi uses column major storage. Natural way to implement fast reshape is to reshape matrix without reordering data. 

Also, probably CasADi  behavior consistent with matlab.

вторник, 15 июня 2021 г. в 18:26:34 UTC+3, harzer...@gmail.com:

[Joel Andersson]

Kirill's answer is correct. CasADi uses column major storage, like MATLAB or Fortran. Numpy can actually use either row-major or column major storage, but the default is row-major. You should be able to get column-major by setting the "order" parameter to "F":

numpy.reshape(a, newshape, order='F')

Joel