Tf.linalg.eig & np.linalg.eig

Why the eigenvectors output is different for tensorflow and numpy?

Just the eigenvectors or also the eigenvalues?
As far as I know, they’re different in the output dtype, but should be equal in the values.
tf.linalg.eig() returns a dtype=tf.complex64, while
np.linalg.eig() depends on your input np.dtype.

You can try to pass your input as dtype=np.complex64 for np.linalg.eig(), then they shouldn’t be different anymore (both as complex numbers).