We implement our models in PyTorch 1.0.0, thus the following packages need to be installed:
pip intall ...
First, download dataset.tar.gz using http:https://bit.ly/2VL9lg6 and put in the root folder of this project. Then unpack it by running:
tar -xzf dataset.tar.gz
This will create the dataset folder, with the following files:
dict.pt # pickle archive of behavior_clone/inst_dict.py, which contains the words and instructions dictionaries.
dev.json # a json dump of a small training dataset useful for debugging.
train.json # a json dump of a large training dataset.
val.json # a json dump of a validation dataset.