AttributeError: 'reshape' object has no attribute '_lazy_canonical_form'. Did you mean: 'canonical_form'?

[raff0722]

AttributeError: 'reshape' object has no attribute '_lazy_canonical_form'. Did you mean: 'canonical_form'?

[SteveDiamond]

You have to solve the problem.

[SteveDiamond]

I was able to reproduce the issue.

[PTNobel]

import cvxpy as cp
import numpy as np
d = 10
np.random.seed(0)
A = np.random.randn(5, 10)
B = np.random.randn(4, 10)
C = np.random.randn(3, 10)
x = cp.Variable(d)
objective = cp.Minimize(cp.sum_squares(A@x) + cp.mixed_norm(cp.hstack([B @ x, C @ x, np.ones(d)]), 1, 2))

I was unable to reproduce the issue with this script. Is there any info I'm missing?

[SteveDiamond]

@raff0722 you're hitting this bug #1963
In your problem mixed_norm isn't doing anything different than regular norm because the hstack output is 1D. Possibly you didn't mean for it to be 1D. We follow NumPy conventions for hstack, vstack, concatenate, etc.

[PTNobel]

Here's the minimized script:

import cvxpy as cp
import numpy as np
d = 10
objective = cp.Minimize(0)
constr = [cp.mixed_norm(cp.hstack([np.ones(d)]), 1, 2) <= 1]
cp.Problem(objective, constr).solve()

It is worth at least fixing its error message.

[raff0722]

@raff0722 you're hitting this bug #1963 In your problem mixed_norm isn't doing anything different than regular norm because the hstack output is 1D. Possibly you didn't mean for it to be 1D. We follow NumPy conventions for hstack, vstack, concatenate, etc.

Yes, that was the issue. Thank you! I expected a matrix with d rows and 3 columns. Now I used bmat as follows

cp.mixed_norm( cp.bmat([ [B@x], [C@x], [np.ones(d)] ), 1, 2 )

which works.

However, I find the doc about bmat a bit confusing. It says

Takes a list of lists. Each internal list is stacked horizontally. The internal lists are stacked vertically.

Morevoer, based on Meaning in

I would expect that I have to transpose the matrix, i.e., use

cp.mixed_norm( cp.bmat( [ [B@x], [C@x], [np.ones(d)] ] ).T, 1, 2 )

which however gave the wrong result.