search

kevinduh / san_mrc

Stochastic Answer Networks (SAN) for Machine Reading Comprehension
BSD 3-Clause "New" or "Revised" License
148 stars 47 forks source link

About the default model settings #8

Closed jihunchoi closed 5 years ago

jihunchoi commented 5 years ago

Hi, thank you for releasing the code. As far as I understood, the default configuration sets decoder_mem_type to 0, and thus it does not use the stochastic answer network architecture and decoder_mem_drop_p is ignored. The default setting yields a fairly good performance of EM 76.3 and F1 84.1, but can you share the arguments used in experiments for further exploration? Thanks!

namisan commented 5 years ago

I optimized the model arch recently, the baseline model can obtain a good performance. If you set decoder_mem_type=1 and decoder_mem_drop_p=0.2, it should obtain 76.8/84.6 (EM/F1).