You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When using AutoLinearOperator with well_posed=True and no tags specifying some structure, the LU solver is used.
LU wraps the jax.scipy implementation and raises an error if the output has NaN or +/- inf values. However, jsp.linalg.lu_solve will not necessarily return a vector with such values for an ill-posed problem, unless one uses double-precision: jax-ml/jax#23626
If there is no fix in JAX itself, then we could check if the solution is valid and throw an error if it's not.
I am using SVD for that particular problem anyway, but wanted to note this here since AutoLinearSolver promises to raise an error for ill-posed problems.
The text was updated successfully, but these errors were encountered:
For this one I think let's aim to get this fixed upstream in JAX itself. I think it'd be really hard for us to work around this issue here in a reliable way. I'm following the discussion there; let me know if you need any input. :)
When using
AutoLinearOperator
withwell_posed=True
and no tags specifying some structure, the LU solver is used.LU wraps the
jax.scipy
implementation and raises an error if the output hasNaN
or+/- inf
values. However,jsp.linalg.lu_solve
will not necessarily return a vector with such values for an ill-posed problem, unless one uses double-precision: jax-ml/jax#23626If there is no fix in JAX itself, then we could check if the solution is valid and throw an error if it's not.
I am using SVD for that particular problem anyway, but wanted to note this here since
AutoLinearSolver
promises to raise an error for ill-posed problems.The text was updated successfully, but these errors were encountered: