fg-gating
fg-gating copied to clipboard
Fine-grained Gating for Reading Comprehension
Words or Characters? Fine-grained Gating for Reading Comprehension
Intro
This is an implementation of the paper
Words or Characters? Fine-grained Gating for Reading Comprehension
Zhilin Yang, Bhuwan Dhingra, Ye Yuan, Junjie Hu, William W. Cohen, Ruslan Salakhutdinov
ICLR 2017
SOTA results on Children's Book Test (CBT) and Who-Did-What (WDW)
Get data
Download and extract the preprocessed data.
wget http://kimi.ml.cmu.edu/fg_data/cbtcn.tar
tar -xvf cbtcn.tar
wget http://kimi.ml.cmu.edu/fg_data/cbtne.tar
tar -xvf cbtne.tar
wget http://kimi.ml.cmu.edu/fg_data/wdw.tar.gz
tar -xvzf wdw.tar.gz
mkdir wdw_relaxed
cd wdw_relaxed
wget http://kimi.ml.cmu.edu/fg_data/wdw_relaxed/data.tgz
tar -xvzf data.tgz
Requirements
Lasagne + Theano. Python 2.7.
Install Lasagne and Theano with the instructions here: https://github.com/Lasagne/Lasagne#installation
Run the Models
CBTCN
python run.py --dropout 0.4 --dataset cbtcn --seed 1
CBTNE
python run.py --dropout 0.4 --dataset cbtne --seed 31
WDW
python run.py --dropout 0.3 --dataset wdw --seed 11
WDW Relaxed
python run.py --dropout 0.3 --dataset wdw_relaxed --seed 51