@edyoshikun it seems like my on-the-fly affine scaling implementation was buggy (I saw your [::-1]). I'm fairly certain the bugs were caused by np.linalg.eig returning unordered eigenvalues .
Here I'm suggesting a much simpler (and tested) version. Note the PR is against replace_zstep_global.
@edyoshikun it seems like my on-the-fly affine scaling implementation was buggy (I saw your [::-1]). I'm fairly certain the bugs were caused by
np.linalg.eigreturning unordered eigenvalues .Here I'm suggesting a much simpler (and tested) version. Note the PR is against
replace_zstep_global.