It can be useful to expose the eigen_tol parameter in spectral embedding, so that the user can tune the value based on their needs. sklearn also does this.