def _log_prob_shape(dist, x_size=torch.Size()):
event_dims = len(dist.event_shape)
expected_shape = broadcast_shape(dist.shape(), x_size, strict=True)
if event_dims > 0:
expected_shape = expected_shape[:-event_dims]
return expected_shape
def _log_prob_shape(dist, x_size=torch.Size()):
event_dims = len(dist.event_shape)
expected_shape = broadcast_shape(dist.shape(), x_size, strict=True)
if event_dims > 0:
expected_shape = expected_shape[:-event_dims]
return expected_shape