Importance of a Search Strategy in Neural Dialogue Modelling
This repo provides code, trained models, run scripts and human evaluation transcripts for our work on different search strategies for neural dialogue models.
All the training and human evaluation was done using ParlAI framework. ParlAI is being actively developed and we do not confirm our code to be working with master branch code.
It should be working using ParlAI commit
We plan to include scorer functionality together with scorer in main ParlAI repo in the near future.
seq2seq_on_steroids agent please do the following:
git clone firstname.lastname@example.org:facebookresearch/ParlAI.git git clone email@example.com:uralik/beamybeam.git cp -r beamybeam/parlai_external ParlAI/ cd ParlAI; python setup.py develop
After that you should be able to import SteroidSeq2seqAgent using this command:
from parlai_external.agents.seq2seq_on_steroids.seq2seq_on_steroids import SteroidSeq2seqAgent from parlai_external.agents.seq2seq_on_steroids.modules import SteroidSeq2seq
This is the model used for all experiments in the paper. Corresponding
.opt file provides
all hyperparameters which were used during the training.
Archive contains typical set of files needed in ParlAI to do any kinds of further tasks. Please see ParlAI docs for further details.
To make a quick check to verify your model is running you can use eval script after paths adjustments.
Please use the following bib if you wish to cite our work: