Skip to content

glin93/HMNs

Repository files navigation

Introduction

Thanks to the great work in Mem2seq, we code is built upon Mem2seq. bAbI tasks can be found here and DSTC 2 data can be found here.

Requirements

pip install requirements.txt

Change permision for BLEU evaluation script

chmod +777 tools/multi-bleu.perl

Commands to Reproduce HMNs Results HMNs

If you want to run on GPUs, please modify the variable USE_CUDA(in utils/config.py) to be Ture, or set it to be False when you want to run it on CPU.

  1. bAbI-3
python multi_mem_train.py -lr=0.001 -layer=1 -hdd=256 -dr=0.1 -dec=Mem2Seq -bsz=64 -ds=babi -t=3 -kb-layer=1 -enbirnn -debirnn
  1. bAbI-4
python multi_mem_train.py -lr=0.001 -layer=1 -hdd=256 -dr=0.2 -dec=Mem2Seq -bsz=64 -ds=babi -t=4 -kb-layer=1 -enbirnn -debirnn
  1. bAbI-5
python multi_mem_train.py -lr=0.001 -layer=1 -hdd=256 -dr=0.1 -dec=LuongSeqToSeq -bsz=64 -ds=babi -t=5  -kb-layer=1 -enbirnn -debirnn
  1. DSTC2
python multi_mem_train.py -lr=0.001 -layer=1 -hdd=128 -dr=0.1 -dec=Mem2Seq -bsz=64 -ds=babi -t=6 -kb-layer=1 -enbirnn -debirnn  -load-limits=15 
  1. Kek Value Dataset
python multi_mem_train.py -lr=0.001 -layer=3 -hdd=256 -dr=0.1 -dec=Mem2Seq -bsz=64 -ds=kvr -t= -kb-layer=3 -enbirnn -debirnn 

About

Task-Oriented Conversation Generation Using Heterogeneous Memory Networks

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published