Explaining hidden states in recurrent networks for classification tasks with recurrent attention model
