ENH: add models module with models of acoustic communication

[NickleDave]

To go beyond bioacoustics and to actually serve as a core package for acoustic communication researchers, we need a models module that contains reference implementations of models of animal acoustic communication and behavior.

This is a meta issue to keep track of what that module might look like. Please feel free to comment and add to the discussion.

My initial thoughts:

• scope: "scikit-learn-sized" models, mainly written with Python + array-oriented numpy -- neural network models can live in vak
• how to group? by research area or by model type/framework?
  • any schema will be imperfect
  • research area would be something like: vocal learning, cultural evolution, syntax, evolution
  • model type would be something like:
    • vocal learning
      • reinforcement learning
        • Troyer
        • Seung
      • imitation learning
        • De Freitas
      • babbling
        • models that Sita cites in review
    • cultural evolution
      • agent-based models
        • Youngblood
        • Creanza
    • syntax
      • context-free grammars, etc.
        • Sainburg
        • Beckers
      • probabilistic-suffix tree
        • Markowtiz, canary
      • markov
        • Jin, Bengalese finch
    • evolution
      • game theory
      • Matt Spike models
    • collective behavior
    • social behavior
      • network models
        • Stowell: https://github.com/danstowell/callnets-glm
    • production: speech is out of scope, there are plenty of packages for this already
      • source-filter
      • Mindlin model
    • rhythm
      • Troyer et al: https://web.archive.org/web/20220120153928/https://www.utsa.edu/troyerlab/software.html
      • Ravignani et al

🤔 Maybe there's not enough overlap and it's ok to group by research area

[yardencsGitHub]

Seems to me that there are two labels: Types of models (e.g. graphical probabilistic models, mechanistic models ..), and usage (syntax, development .. ). It's not a 1-to-1 mapping.

[NickleDave]

Thanks @yardencsGitHub, agreed. We probably want something more like tags then a hierarchical structure. So we might structure the code base one way (say, math framework of model) but then provide multiple ways to view this in the docs (either by type of model or usage)

[NickleDave]

Nice to have would be bare-bones models like first order Markov + descriptive stats from NLP like n-grams