search

NVlabs / FPSci

Aim Training Experiments
Other
67 stars 23 forks source link

Add adaptive stimulus #420

Open bboudaoud-nv opened 1 year ago

bboudaoud-nv commented 1 year ago

This branch adds support for adapting FPSci stimulus at runtime using external scripting. This model is further detailed in documentation updates to experimentConfigReadme.md.

A sample should be added to demonstrate the concept before we mark this branch ready for review.

The current approach is not heavily tested for mixing adaptive and constant tasks in a single session in FPSci as this is not the core use case. Theoretically the approach should work, but there may be additional issues.

We should think about the upsides/downsides of adopting a scripting model like this to run fixed count (constant) tasks as well as fully adaptive ones.

bboudaoud-nv commented 1 year ago

There are a few topics we should discuss prior to merging this PR, examples include:

  1. The possibility of the adaptation script having the results database file open blocking FPSci from writing new results/dropping data.
  2. The challenges of relying on/synchronizing results files and system state, particularly for multi-user runs of an experiment on a single machine.
  3. The current progress mechanism and if/whether we want to modify it at all to (1) report trial level updates and (2) report overall session-level progress instead of just the current task.
  4. Are there any (new) feedback string values we'd like to include in FPSci to help guide subjects through adaptive stimulus trials?