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
where the eigenvectors are different from those returned by other BLAS implementations. In the case of scipy, which uses evr by default instead of evd, returns the correct eigenvectors; scipy.linalg.eigh(a) returns
To reproduce, i ran
conda install -c conda-forge numpy "libblas=*=*accelerate"
in a new environment. Then,
returns
where the eigenvectors are different from those returned by other BLAS implementations. In the case of
scipy
, which usesevr
by default instead ofevd
, returns the correct eigenvectors;scipy.linalg.eigh(a)
returnsBy specifying the LAPACK driver with
scipy.linalg.eigh(a, driver='evd')
, I again obtain the same result as numpy.After some testing with the LAPACK calls
scipy.linalg.lapack.zheevr
andscipy.linalg.lapack.zheevd
,zheevd
does not seem to return correct values.Environment (
conda list
):Details about
conda
and system (conda info
):The text was updated successfully, but these errors were encountered: