Reorganize Aqualab Features

[LswaN58]

The Problem

We have had a lot of development of Aqualab features over the course of several years, for a wide variety of projects, across a wide variety of interns, and even across very different versions of the OGD core infrastructure and best practices.

As a result, the feature extractor set has become disorganized and difficult to keep track of. For example, we have a few different approaches to time counting across somewhere around a dozen different time-based features. Some approaches work better than others, and it becomes nearly impossible to remember which time features are "good" and which are "that thing that we needed for one project but shouldn't use anymore because it has weird assumptions." Thus, the time has come to review the feature set, identify which features are still worth including in the base feature set, which need to be rewritten to better follow our best practices, and which can remain as they are.

Time Features

The Old Way

The feature set for time-based features as of November 2024 was roughly as follows:

1. ActiveTime
2. AverageSessionTime
3. JobActiveTime – per job
4. JobArgumentation – per Time - per job
5. JobCompletionTime – per job
6. JobDiveTime – per Active/ per job
7. JobExperimentation – per Time – per job
8. JobModeling – per Time – per job
9. JobPlayTime
10. SessionDuration
11. SyncCompletionTime – per job
12. TotalArcticTime
13. TotalArgumentationTime – per Seconds/Active-Seconds/Active/Idle-Seconds/Idle
14. TotalBayouTime
15. TotalCoralTime
16. TotalDiveTime – per Active-Seconds/Idle/Seconds/Idle-Seconds
17. TotalExperimentationTime -per Seconds/Active/Idle-Seconds/Active-Seconds
18. TotalKelpTime
19. TotalModelingTime – per Idle-Seconds/Idle/Active/Active-Seconds/Seconds
20. TotalPlayTime – per Active-Seconds/Seconds/Idle/Active/Idle-Seconds
21. UserAvgActiveTime
22. UserAvgSessionDuration
23. UserTotalSessionDuration

Clearly, time-based features needs a lot of work.

The New Way

The proposed new set of features:

• Total time features - these three are the same thing, at different granularities:
  • Total Session Time
  • Total Player Time
  • Total Everybody Time
• Per-job time at player level
• Per-region times:
  • TotalKelpTime
  • TotalCoralTime
  • TotalArcticTime
  • TotalBayouTime
• Per-mechanic times:
  • Total times:
    • TotalModelingTime
    • TotalExperimentationTime
    • TotalDiveTime
    • TotalArgumentationTime
  • Per-job times:
    • JobModelingTime
    • JobExperimentationTime
    • JobDiveTime
    • JobArgumentationTime
• Population-level averages - tentative, maybe do:
  • Average session time
  • Average player time
  • Per-job average time
• Other stuff we could do:
  • Time proportions (e.g. spent 20% of job5 time in modeling)
  • Time percentiles (e.g. spent more time in job5 than 75% of others)

Where we want to get, big-picture

• All time features should have total, active, and idle times reported

Other Features

To-do. Need a review of features to get list of what exists, and where we want to go.