Extensions to Hybrid Code Networks for FAIR Dialog Dataset

Jiyeon Ham, Soohyun Lim, Kyeng-Hun Lee, Kee-Eung Kim

Getting started

Version info

• Python 2.7
• Keras 2.1.4

Prerequisite

• Download dataset from https://github.com/perezjln/dstc6-goal-oriented-end-to-end
• Make data directory and unzip the dataset
• Make weight directory

Then tree veiw should be shown as:

├─data
│  ├─dataset-E2E-goal-oriented
│  └─dataset-E2E-goal-oriented-test-v1.0
│      ├─tst1
│      ├─tst2
│      ├─tst3
│      └─tst4
├─scripts
└─weight

Training

run scripts/main.py with the following arguments:

• -t: train
• -et: entity tracking module
• -as: action selector module
• -eo: entity output module
• -ts: task number to train (only used for action selector module)

Train entity tracking module

$ python scripts/main.py -t -et

Train action selector module for task 1

$ python scripts/main.py -t -as -ts 1

Train entity output module

$ python scripts/main.py -t -eo

Predict

run scripts/main.py with the following arguments:

• -us: test data with unseen slot
• -oov: test data with out-of-vocabulary knowledge base
• -ts: task number to predict

Predict for task 1 with unseen slot and out-of-vocabulary

$ python main.py -us -oov -ts 1