Method jvp

tensorflow/python/eager/forwardprop.py:192–211

Fetches the Jacobian-vector product computed for `target`. Note that this function performs no computation, and simply looks up a JVP that was already computed (unlike backprop using a `tf.GradientTape`, where the computation happens on the call to `tape.gradient`). Args:

(self, target)

      pywrap_tensorflow.TFE_Py_ForwardAccumulatorWatch(self._accumulator, t, g)

  def jvp(self, target):
    """Fetches the Jacobian-vector product computed for `target`.

    Note that this function performs no computation, and simply looks up a
    JVP that was already computed (unlike backprop using a
    `tf.GradientTape`, where the computation happens on the call to
    `tape.gradient`).

    Args:
      target: A watched Tensor or structure of Tensors to fetch the JVPs for.

    Returns:
      Tensors with the same shapes and dtypes as `target`, or None if no JVP
      is available.
    """
    if self._accumulator is None:
      raise ValueError("Called jvp() without first tracing anything.")
    return nest.map_structure(
        functools.partial(pywrap_tensorflow.TFE_Py_ForwardAccumulatorJVP,
                          self._accumulator), target)