• not a good choice for tasks that have very high dimensional sparse features like text classification, where linear models can provide efficient training and fast accurate prediction