relative_errors
===============

.. py:function:: t3toolbox.backend.optimizer_display.relative_errors(residual_block_sumsq, data_block_sumsq)

   .. code-block:: python

      def relative_errors(
              residual_block_sumsq: NDArray,   # kind.block_sumsq(residual, n_w), shape (n_mode, n_order)
              data_block_sumsq:     NDArray,   # kind.block_sumsq(data, n_w),     shape (n_mode, n_order)
      ) -> NDArray:                            # sqrt(num/den) per block; NaN where the data block norm is 0


   Per-``(mode, order)`` relative error ``‖r_ij‖ / ‖y_ij‖ = sqrt(block_sumsq(r) / block_sumsq(y))``.
   A zero-norm data block (no signal in that block) is undefined -> ``NaN`` (rendered ``—``).