facebookresearch / Mephisto

A suite of tools for managing crowdsourcing tasks from the inception through to data packaging for research use.
https://mephisto.ai/
MIT License
303 stars 76 forks source link

[Refactor] Move all data file manipulation to MephistoDB #567

Open JackUrb opened 3 years ago

JackUrb commented 3 years ago

Overview

At the moment a number of the classes we use in Mephisto for storing data or preparing task information are managed by classes outside of the MephistoDB. This prevents us from reaching our ideal state of fully abstracting out the storage backend from the data being collected. While not necessarily high-priority to resolve at the moment, this PR documents the locations where this abstraction is leaking.

Ultimately, many individuals' tasks may find it better to continue doing direct writes in AgentState's for instance, but we should strive for clean abstractions internally.

Implementation plan

Etesam913 commented 2 years ago

This should be closed right?

JackUrb commented 2 years ago

The core of it is now implemented, but some code in the Mephisto codebase direct writes to and reads from files even when a db is accessible. I'll close this once those are all cleaned up.