SharedGeometry.frame
====================

.. py:method:: t3toolbox.shared_geometry.SharedGeometry.frame(x)

   .. code-block:: python

      def frame(self, x):


   The base geometry's frame at ``x`` (ragged ``TuckerTensorTrain`` -> ``T3Frame``;
   uniform ``UniformTuckerTensorTrain`` -> ``UT3Frame``).

   **An untied ``x`` is tied first, silently**, by the per-group mean
   (:py:func:`~t3toolbox.backend.sharing.t3_tie_tucker_factors` / its uniform twin) -- the frame of
   a shared geometry describes a point on the shared set, so entering the format is this method's
   job rather than the caller's. An already-tied point is a bitwise fixed point, so the ordinary
   path is unchanged. This is what makes an untied initial guess a non-event for the frontend optimizers (which build their frames through this wrapper),
   and it also absorbs the slow drift a long run of low-precision first-order steps can produce.