Personalized Dialog Generation with Commonsense

Huggingface cleaned dataset

wget https://s3.amazonaws.com/datasets.huggingface.co/personachat/personachat_self_original.json

(Optional) Personachat Download: http://parl.ai/downloads/personachat/personachat.tgz

Clone repo from: https://github.com/atcbosselut/comet-commonsense From repo:

Then run the setup scripts to acquire the pretrained model files from OpenAI, as well as the ATOMIC and ConceptNet datasets

bash scripts/setup/get_atomic_data.sh
bash scripts/setup/get_conceptnet_data.sh
bash scripts/setup/get_model_files.sh

Make sure you have all the requirements mentioned here in README: https://github.com/atcbosselut/comet-commonsense

Make preprocessed data loader for ATOMIC and CONCEPTNETS

python scripts/data/make_atomic_data_loader.py
python scripts/data/make_conceptnet_data_loader.pypython scripts/data/make_atomic_data_loader.py

Pretrined models can be downloaded from here: https://drive.google.com/open?id=1FccEsYPUHnjzmX-Y5vjCBeyRt1pLo8FB

Unzip the file: tar -xvzf pretrained_models.tar.gz

Play with COMeT completions here: python scripts/interactive/atomic_single_example.py --model_file pretrained_models/atomic_pretrained_model.pickle

Choose all as effect type. Other options as follow:

all - compute the output for all effect types {{oEffect, oReact, oWant, xAttr, xEffect, xIntent, xNeed, xReact, xWant}}
oEffect - generate the effect of the event on participants other than PersonX
oReact - generate the reactions of participants other than PersonX to the event
oEffect - generate what participants other than PersonX may want after the event

Choose beam-5 as decoding algorithm. Other options as follow:

greedy
beam-# where # is the beam size
topk-# where # is k

We will change this code to be able take an input json and produce an output json with all expansions.