Training details for bAbI-10k dataset

[taey16]

Hi, @vinhkhuc

Courtesy of your repo. we got following learning curves for bAbI 1k dataset.


We think above results are almost the same as the MemN2N paper.

But, we didn't get expected results for bAbI-10K dataset.
As the author said, we set nepoch=20, anneals learning rate ratio of a half, and LS with initial learning rate=0.005, however, our results got reached at around 30% error. We think we need much hypo-parameter tuning. If you have time, give us comments or idea.

thanks,

taey16