Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

BUG: pt.linalg.solve returns incorrect results when mode = "NUMBA" #422

Open
jessegrabowski opened this issue Aug 25, 2023 · 3 comments · May be fixed by #1146
Open

BUG: pt.linalg.solve returns incorrect results when mode = "NUMBA" #422

jessegrabowski opened this issue Aug 25, 2023 · 3 comments · May be fixed by #1146
Assignees
@jessegrabowski
Labels
bug Something isn't working linalg Linear algebra numba

Comments

@jessegrabowski
Copy link
Member

jessegrabowski commented Aug 25, 2023 •
edited
Loading

Describe the issue:

This is identical to #382. When assume_a != 'gen' in pt.linalg.solve, the solve_triangular function is incorrectly invoked, and the resulting output is not correct.

Reproducable code example:

import pytensor
import pytensor.tensor as pt
import numpy as np

X = pt.dmatrix('X')
b = pt.dmatrix('b')
y1 = pt.linalg.solve(X, b, assume_a='gen')
y2 = pt.linalg.solve(X, b, assume_a='sym')

f1 = pytensor.function([X, b], y1)
f2 = pytensor.function([X, b], y2)
f1_nb = pytensor.function([X, b], y1, mode='NUMBA')
f2_nb = pytensor.function([X, b], y2, mode='NUMBA')

X_sym = np.random.normal(size=(3, 3))
X_sym = X_sym @ X_sym.T
X_inv_1 = f1(X_sym, np.eye(3))
X_inv_2 = f2(X_sym, np.eye(3))
X_inv_1_nb = f1_nb(X_sym, np.eye(3))
X_inv_2_nb = f2_nb(X_sym, np.eye(3))


# Passes, C backend, assume_a = 'gen'
np.testing.assert_allclose(X_inv_1 @ X_sym, np.eye(3), atol=1e-12)

# Passes, C backend, assume_a = 'sym'
np.testing.assert_allclose(X_inv_2 @ X_sym, np.eye(3), atol=1e-12)

# Passes, Numba backend, assume_a = 'gen'
np.testing.assert_allclose(X_inv_1_nb @ X_sym, np.eye(3), atol=1e-12)

# Fails, Numba backend, assuma_a = 'sym'
np.testing.assert_allclose(X_inv_2_nb @ X_sym, np.eye(3), atol=1e-12)

Error message:

AssertionError: 
Not equal to tolerance rtol=1e-07, atol=1e-12

Mismatched elements: 6 / 9 (66.7%)
Max absolute difference: 0.67369019
Max relative difference: 0.52399333
 x: array([[ 4.760067e-01, -2.459194e-01,  9.939378e-18],
       [-3.325732e-01,  6.636944e-01,  1.353323e-18],
       [-1.317192e-01,  6.736902e-01,  1.000000e+00]])
 y: array([[1., 0., 0.],
       [0., 1., 0.],
       [0., 0., 1.]])

PyTensor version information:

Pytensor version 2.11.1

Context for the issue:

No response
@jessegrabowski jessegrabowski added bug Something isn't working numba linalg Linear algebra labels Aug 25, 2023
@jessegrabowski jessegrabowski self-assigned this Aug 25, 2023
@ricardoV94
Copy link
Member

@jessegrabowski Is this one resolved?

@jessegrabowski
Copy link
Member Author

No. I am working on a numba overload for solve but it's more work than I expected. Maybe in the meantime I could do a temporary quick fix that either errors if assume_a is not "gen", or silently force it to be?

@ricardoV94
Copy link
Member

Yes you can treat it as if it was gen always while we don't have the alternatives working

@jessegrabowski jessegrabowski linked a pull request Dec 31, 2024 that will close this issue
Open
16 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@jessegrabowski jessegrabowski

Labels
bug Something isn't working linalg Linear algebra numba
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Add LAPACK overloads for all variants of pt.linalg.solve jessegrabowski/pytensor
2 participants
@ricardoV94 @jessegrabowski