CyLP
CyLP copied to clipboard
coin-or
Use of numpy.matrix not recommended
The use of numpy.matrix is discuraged since numpy 1.15
Note It is no longer recommended to use this class, even for linear algebra. Instead use regular arrays. The class may be removed in the future.
Using regular arrays gives some problems, like in the first example:
import numpy as np
from cylp.cy import CyClpSimplex
from cylp.py.modeling.CyLPModel import CyLPArray
s = CyClpSimplex()
# Add variables
x = s.addVariable('x', (3, 1)) # should be 2-d now
y = s.addVariable('y', (2, 1)) # should be 2-d now
# Create coefficients and bounds
A = np.array([[1., 2., 0], [1., 0, 1.]]) # use array
B = np.array([[1., 0, 0], [0, 0, 1.]]) # use array
D = np.array([[1., 2.], [0, 1]]) # use array
a = CyLPArray([5, 2.5])
b = CyLPArray([4.2, 3])
x_u = CyLPArray([2., 3.5])
# Add constraints
s += A @ x <= a # use matmul operator
s += 2 <= B @ x + D @ y <= b # use matmul operator
s += y >= 0
s += 1.1 <= x[1:3] <= x_u
# Set the objective function
c = CyLPArray([1., -2., 3.])
s.objective = c @ x + 2 * y.sum() # use matmul operator
# Solve using primal Simplex
s.primal()
print(s.primalVariableSolution['x'])
Traceback (most recent call last):
File "/path/to/example.py", line 20, in <module>
s += A @ x <= a
ValueError: matmul: Input operand 1 does not have enough dimensions (has 0, gufunc core with signature (n?,k),(k,m?)->(n?,m?) requires 1)
Process finished with exit code 1
Ah, OK, I did not realize this. It seems to me that using the matmul operator is probably not going to work in any case, since this operation is not defined for the classes within CyLP. Do you have the same issue if you use the regular mul operator?
Hmm, using the mul operator also gives an error, but a different one. It's not exactly clear to me what's going on. It doesn't appear that np.matrix is actually being used anywhere underneath, just in the example, to construct the matrix. The matrices are stored internally using scipy's sparse matrix classes. Somehow, there is a shape mismatch.
I found that np.asmatrix solves the problem
As simple as:
A = np.asmatrix([[1., 2., 0],[1., 0, 1.]])
The NumPy documentation does not contain the warning about future use: https://numpy.org/doc/stable/reference/generated/numpy.asmatrix.html
That should not be the case as you still work with the matrix class, while you are encouraged not to, as the class may be removed in the future:
It is no longer recommended to use this class, even for linear algebra. Instead use regular arrays. The class may be removed in the future.
I so guess the documentation could be improved.
From the source:
@set_module('numpy')
def asmatrix(data, dtype=None):
"""
<docs left out for brevity>
"""
return matrix(data, dtype=dtype, copy=False)
@akeeman you are absolutely right I should have had to check it out myself
For future similar questions, I'm using this workaround:
from scipy import sparse
A= sparse.csr_matrix(A)