Finished: Attentive Model from Teaching machine to read and comprehend.
With batch size=128, vocab size=50000, learning rate 5e-5, RMSProp optimizer, the model achieves 55% accuracy on test set.
To train the model (First preprocess the data using tools in utils/data_utils.py):
cd ./attentive-reader
./main.py --learning_rate 0.00005 --vocab_size 50000 --optim RMS --attention concat --activation tanh
Current: playing with structures, trying different attention mechanism
My work is partly based on carpedm20's Code.
============= Other Model: in construction