Migrate Han's RL models to this repo

[hanhou]

Related repos

• Original repo: https://github.com/hanhou/Dynamic-Foraging
• How I organized the model tables: https://github.com/hanhou/map-ephys/blob/de8e5e2c6c5aa9c2dc0df1a1b82b492554e21f9c/pipeline/foraging_model.py#L151-L152
• Stefano's refactoring: https://codeocean.allenneuraldynamics.org/capsule/3729985/tree

[hanhou]

• developing in this capsule https://codeocean.allenneuraldynamics.org/capsule/1652248/tree

[hanhou]

Things to improve over Stefano's refactoring:

1. Even better class architecture
2. Model cross-validation and model comparison
3. Use pydantic to validate Params
4. Add more models (all in my previous implementation)
5. Code cleaning
6. Add unittest
7. Improve documentation

[hanhou]

Steps:

• [x] First make .perform() and .fit() of Hattori 2019 work: #23
• [x] Add cross_validation #24
• [x] Add more model variants
  • [x] Add dynamic generation of ParamsModel #27
  • [x] Make sure choice_kernel is working #27
  • [x] Add epsilon-greedy #28
  • [x] Refactor classes #30
  • [x] Add loss counting #31
• [x] Add model presets #32

• [x] Add model playground to the Streamlit app https://github.com/AllenNeuralDynamics/foraging-behavior-browser/pull/87

• [ ] Put it in on-pipeline analysis

• [ ] Add model comparison class

• [ ] Add ignore trials
  • [ ] remove ignore trials if don't fit ignore

  • [ ] add fitting ignore trials

[hanhou]

This PR is almostly done. Will create new issues for remaining points.