Skip to content
question answering, reading comprehension toolkit
Python Shell
Branch: master
Clone or download
lixinsu Feature/fix debug mode (#40)
* #35 add rl-based F1-reward loss

* speedup data load in debug mode
Latest commit 37fcb79 Jun 17, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
experiments/squadv1/bert #33 add bert code and config (#34) Jun 14, 2019
reader Feature/fix debug mode (#40) Jun 17, 2019
runs Add label smooth Apr 2, 2019
scripts/reader Feature/fix debug mode (#40) Jun 17, 2019
LICENSE Create LICENSE Aug 13, 2018
README.md Update README.md Oct 16, 2018
download.sh Update download.sh Oct 15, 2018
search_parameter.py New way to process config Apr 24, 2019

README.md

Introduction

The RCZoo project is a toolkit for reading comprehension model. It contains the PyTorch reimplement of multiple reading comprehension models

Usage

  • run sh download.sh to download the dataset and the glove embeddings.
  • run sh runs/train_squad.sh [bidaf|drqa|slqa|fusionnet|docqa] to start the train process. (Check the xxx.sh scripts before run, as the preprocessing only need to be executed once)

Dependencies

python 3.5
Pytorch 0.4
tqdm

performance

We train each model on train set for 40 epoch, and report the best performance on dev set.

Model Exact Match F1 EM(+ELMo) F1(+ELMo)
Rnet 69.25 78.97
BiDAF 70.47 79.90 73.04 81.48
documentqa 71.47 80.84
DrQA 68.39 77.90
QAnet ... ...
SLQA 67.09 76.67
FusionNet 68.27 77.79

Current progress

Rnet

BiDAF

documentqa

DrQA

QAnet

SLQA

FusionNet

acknowledgement

some code are borrowed from DrQA, a cool project about reading comprehension.

You can’t perform that action at this time.