Method get_auxiliary_parameters

caffe2/python/optimizer.py:252–270

Returns a list of auxiliary parameters. Returns: aux_params: A namedtuple, AuxParams. aux_params.local stores a list of blobs. Each blob is a local auxiliary parameter. A local auxiliary parameter is a parameter in parallel to a learning rate

(self)

            return grad

    def get_auxiliary_parameters(self):
        """Returns a list of auxiliary parameters.

        Returns:
            aux_params: A namedtuple, AuxParams.

            aux_params.local stores a list of blobs. Each blob is a local
            auxiliary parameter. A local auxiliary parameter is a parameter in
            parallel to a learning rate parameter. Take adagrad as an example,
            the local auxiliary parameter is the squared sum parameter, because
            every learning rate has a squared sum associated with it.

            aux_params.shared also stores a list of blobs. Each blob is a shared
            auxiliary parameter. A shared auxiliary parameter is a parameter
            that is shared across all the learning rate parameters. Take adam as
            an example, the iteration parameter is a shared parameter, because
            all the learning rates share the same iteration parameter.
        """
        return self._aux_params

    # TODO(xlwang): In transfer learning, parameter initialized from pretrained
    # model might require a different learning rate than otherwise initialized.