How do I run the Code to Generate Text Adventures

[simonhughes22]

The documentation focuses on the logic to train a model on a number of tasks. I am particularly interested in replicating the work in this paper https://arxiv.org/pdf/1911.09194.pdf about generating text adventures. The documentation does not list where the core LIGHT dataset is (only the WILD data), or how to load it and interact with it in a text adventure game. Please advise.

[stephenroller]

@JackUrb do you want to answer this?

You're right. It looks like a bunch of datasets might be missing from our list of tasks. Thanks for bringing that to our attention.

The original LIGHT dataset is light_dialog. parlai display_data -t light_dialog.

[JackUrb]

Hi @simonhughes22 - sorry for the delay here! We've been working on putting all of the LIGHT code and datasets into a repo for organization and reproducibility purposes, and should have code for training such a model available there in coming months.

At the moment, the results of that particular paper are somewhat hard to reproduce, as the underlying model and code we used in that paper have fallen out of sync with ParlAI. You can certainly use the Starspace model available in the current repo alongside the data in the environment.db file that comes along with the light_dialog task to train your own model though.

[github-actions]

This issue has not had activity in 30 days. Please feel free to reopen if you have more issues. You may apply the "never-stale" tag to prevent this from happening.