fixes autowater in warmup curriculum

[micahwoodard]

Fixes autowater in curriculums subsequently fixing warmup bug of block not switching every trial. Also adds minimum reward check into warmup and removes from coupled trial generator

While I was fixing curriculums also added stage names to task logic

closes #110
closes #108

[micahwoodard]

wait for #129 to be merged

[alexpiet]

@micahwoodard Do we have any tools for viewing these curriculums other than reading the JSON files? Its hard to parse

[micahwoodard]

@alexpiet I don't have any. @bruno-f-cruz do you or does scicomp?

[bruno-f-cruz]

Depends on what "viewing these curricula other than reading the JSON files" requires. The following pieces of information come to mind:

• If you want to look at the structure of the curriculum, Jonathan from SciComp wrote a function to export a diagram (https://github.com/AllenNeuralDynamics/aind-behavior-curriculum/blob/494bcc3e124a7c6e2777bdddf63362ee4e21cbb1/src/aind_behavior_curriculum/curriculum_utils.py#L29). This is what I am currently using to generate diagrams in the docs automatically: https://allenneuraldynamics.github.io/Aind.Behavior.VrForaging/curricula_diagrams/depletion/. Def something that can be improved visually.
• If you want to look at curriculum progression, I believe @dbirman is working on something (I cant find it in the data portal, so perhaps still dev stages).
• Alternatively, we have been using http://aind-behavior.corp.alleninstitute.org:8787/ for curriculum management (i.e.: check which stage the animal is in, assign subjects to a curriculum, change the curriculum, etc...).
• For dynamic foraging in particular, I thought the long-term goal was to refactor the app you already have but to ingest the current format instead (I may have misunderstood tho). That MAY be the way to move forward since the platform could have a bespoke visualization tool (which is what i suspect you may be asking for?), and SciComp (e.g. Dan) can still maintain infrastructure that is shared across multiple behavior platforms.
• If you have suggestions on infrastructure that would be generalizable across multiple behavior platforms, I think that should def be brought up in the behavior working group. The whole point of centralizing infrastructure is exactly to facilitate the development of these tools.
• On a more personal note: as a code reviewing tool, I think we should def get used to looking at JSON (or other markup languages) diffs. These formats that are very easy to diff, and LLMs can help interpret them if you need too.

[dbirman]

You can see curriculum stage information in the project views, flip the "color by" dropdown to "curriculum stage"

[dbirman]

Read this more carefully: yes I think we should develop a generic curriculum viewer and stick it in the data portal, I'll spend some time on that. We also have the project pages which work similarly to what Bruno has for VR foraging, but there's no support right now for seeing the "next session" plan because it's stored in dataverse and we don't have a proper API yet for dataverse access (currently you can only access tables directly, which is a level of access that we should not build into the data portal). Bruno also has the custom session figures he generates, and we will slowly catch up to that on the data portal for VR foraging and we can do the same for dynamic foraging.

All that said, I don't think there's any reason for projects/platforms not to maintain their own viewers if they find that useful or want to stay ahead of what I can maintain with the data portal. The data portal will always be a subset of the things you want because it's public facing and has to be a little more intuitive / self-explanatory then whatever bespoke view you might want. My general view with agents working as well as they do is that scientists should build stuff out that they find useful and talk to scientific computing regularly about what they're doing, and when someone wants to move on they should let us know so we can take over the "good" parts of what you built and maintain them long-term.

[micahwoodard]

@alexpiet the json changes here are setting/adding autowater parameters that match the current curriculums here as well as adding a minimum reward requirement for warmup. The json are auto built from the curriculums if that is easier to look at