Method forward

complexnn.py:282–393

(self, inputs)

        assert(xr.size(1) == self.num_features)

    def forward(self, inputs):
        #self._check_input_dim(xr, xi)

        #xr, xi = torch.chunk(inputs,2, axis=self.complex_axis)
        xr, xi = torch.chunk(inputs,2, dim=self.complex_axis)   #for torch_version 1.1.0
        exponential_average_factor = 0.0

        if self.training and self.track_running_stats:
            self.num_batches_tracked += 1
            if self.momentum is None:  # use cumulative moving average
                exponential_average_factor = 1.0 / self.num_batches_tracked.item()
            else:  # use exponential moving average
                exponential_average_factor = self.momentum

        #
        # NOTE: The precise meaning of the "training flag" is:
        #       True:  Normalize using batch   statistics, update running statistics
        #              if they are being collected.
        #       False: Normalize using running statistics, ignore batch   statistics.
        #
        training = self.training or not self.track_running_stats
        redux = [i for i in reversed(range(xr.dim())) if i!=1]
        vdim  = [1] * xr.dim()
        vdim[1] = xr.size(1)

        #
        # Mean M Computation and Centering
        #
        # Includes running mean update if training and running.
        #
        if training:
            Mr, Mi = xr, xi
            for d in redux:
                Mr = Mr.mean(d, keepdim=True)
                Mi = Mi.mean(d, keepdim=True)
            if self.track_running_stats:
                self.RMr.lerp_(Mr.squeeze(), exponential_average_factor)
                self.RMi.lerp_(Mi.squeeze(), exponential_average_factor)
        else:
            Mr = self.RMr.view(vdim)
            Mi = self.RMi.view(vdim)
        xr, xi = xr-Mr, xi-Mi

        #
        # Variance Matrix V Computation
        #
        # Includes epsilon numerical stabilizer/Tikhonov regularizer.
        # Includes running variance update if training and running.
        #
        if training:
            Vrr = xr * xr
            Vri = xr * xi
            Vii = xi * xi
            for d in redux:
                Vrr = Vrr.mean(d, keepdim=True)
                Vri = Vri.mean(d, keepdim=True)
                Vii = Vii.mean(d, keepdim=True)
            if self.track_running_stats: