* Faster LSTM * PEP8 * RNN dropout fix * PEP * PEP * Less code duplication * LSTM benchmark example * PEP * Test implementation modes * Go through Keras backend