Displaying 1 to 1 from 1 results

keras-SRU - Implementation of Simple Recurrent Unit in Keras

  •    Python

No longer a problem to have different input dimension than output dimension. Performance degrades substantially with larger batch sizes (about 6-7% on average over 5 runs) compared to 1 layer LSTM with batch size of 128. However, a multi layer SRU (I've tried with 3 layers), while a bit slower than a 1 layer LSTM, gets around the same score on batch size of 32 or 128.