feature(controller): 4 - put together the components for the upgraded AER Controller

[hollandjg]

Description

Create new Controllers which supersede the SimpleCycle.

This is part of the solution to:

• feature: add asynchronous cycle which can deal with long-running evaluations #238
• feat: executor dispatch for processing-hungry tasks #247
• feat: save partially evaluated cycle to file #248

This builds on:

• feature(controller): 1 - add controller state objects for snapshots and history #299

• feature(controller): 2 - add planner functions to select the next step for the controller #303

• feature(controller): 3 - add "executor collections" to wrap the sklearn-theorists, experimentalist pipelines and experiment runners #304

• resolves feat: improve seeding interface for Cycle #233

Todos:

• Convert __init__.py into a notebook and include in documentation. Can we still use it as doctests?

Type of change:

• New feature (non-breaking change which adds functionality)

Features:

• Added a BaseController (base.py)
  Handles shared logic (how to get the next step in the cycle, for instance) for all the controller classes
• Added a Cycle, derived from the BaseController (cycle.py)
  Has the same interface as the SimpleCycle but uses the Snapshot state from feature(controller): 1 - add controller state objects for snapshots and history #299, the full_cycle_planner from feature(controller): 2 - add planner functions to select the next step for the controller #303, and the default_online_executor_collection from feature(controller): 3 - add "executor collections" to wrap the sklearn-theorists, experimentalist pipelines and experiment runners #304
• Added a Controller, derived from the BaseController (controller.py),
  Offers easier seeding and a history function than the Cycle, and the ability to choose a different planner function. It uses the Hitory state from feature(controller): 1 - add controller state objects for snapshots and history #299, the last_result_kind_planner (as a default) from feature(controller): 2 - add planner functions to select the next step for the controller #303, and the default_online_executor_collection from feature(controller): 3 - add "executor collections" to wrap the sklearn-theorists, experimentalist pipelines and experiment runners #304.
  Resolves feat: improve seeding interface for Cycle #233
• Added documentation showing all the classes in use (__init__.py):
  • simple execution using the Cycle,
  • more complex execution and customization using the Controller and
  • full customization and multiple experimentalists by using the BaseController

[musslick]

This is neat!

[musslick]

I think this looks fantastic. Especially the ability to define your own planner and introduce contingencies about which experimentalist/theorist to use dependent on the history object.

The examples in the init are really well documented and this makes we wonder whether we could supply a python notebook here that we could easily integrate in the documentation (or on Google Collab)? I think this would help users a lot. We could literally use the init.py examples as the notebook. What do you think?

[musslick]

Love it!

[hollandjg]

I think this looks fantastic. Especially the ability to define your own planner and introduce contingencies about which experimentalist/theorist to use dependent on the history object.

The examples in the init are really well documented and this makes we wonder whether we could supply a python notebook here that we could easily integrate in the documentation (or on Google Collab)? I think this would help users a lot. We could literally use the init.py examples as the notebook. What do you think?

Yes, we totally should. I'm not yet sure the best way to do that and integrate the notebook with the documentation though.

[hollandjg]

this is too big. Closing for now, will move this to a separate repo. Thanks for the feedback!