refactor run system API

[Rykath]

Goals

simplify run components by:

• reducing inter-dependencies
• simpler & cleaner API access
• easier unit testing
• better documentation: available options and default values are documented with the API

Major Changes

some run components now have different options

* `Runner` * `run_dir` →`work_dir` (location of run directories, logs and temporary files) - reduce confusion with individual run directories * ... todo

the run configuration uses a different structure

```yaml run: runner: interface: worker: debug: will override "runner.debug" & "worker.debug" when set to "true" ``` To keep some backwards compatibility the `pre` and `post` subsections are also allowed. They are passed to the Worker and should be directly specified there: ```yaml run: pre: template post: numpytxt ``` is allowed but is better written as: ```yaml run: worker: class: command pre: template post: numpytxt ``` The `command` shorthand will be expanded to a `command`-worker ```yaml run: command: ./simulation.py ``` is expanded to: ```yaml run: worker: class: command command: ./simulation.py ``` Finally the following options are set from other sections of the config: * `runner.work_dir` ←`base_config.run_dir` * `interface.size` ← `base_config.ntrain` * `interface.input_config` ← `base_config.input` (generated by `variables`) * `interface.output_config` ← `base_config.output` (generated by `variables`) ---

Minor Changes

• Worker will no longer read the config file, environment variables are used instead
• if the interface fails, backups are no longer checked for
• the default Worker is now the CommandWorker
• custom Workers no longer require Pre- and Postprocessors
• new Component abstract base class makes registering custom components easier
• added unit tests for all run components - increases coverage to >82% (except slurm)
• new searchtype option for simple Active Learning - workaround for #175

Tasks

• [x] Worker follows the same component structure as the others: abstract base class
• [x] Worker does not always create Pre- and Postprocessor
  • the default Worker is the CommandWorker which requires a Pre- and Postprocessor
  • other (custom) Workers work without Pre- and Postprocessor (this means they also create no run directory)
• [x] use new Component abstract base class for all run components
• use arguments instead of config
  • [x] options in init are the same as the config options
  • [x] interfacing with Config classes
  • [x] defaults are now found in the class definitions again
• unit tests
  • [x] Runner- and Worker-Interface
  • [x] Template-Preprocessor and Postprocessors
  • [x] Runner
  • [x] test Slurm-Runner (example & unit test) → coverage of slurm.py: 71% (37/129)
  • [x] Command-Worker
• [x] update documentation
  • [x] added the two example notebooks to the documentation
• [x] API example (notebook)
• [x] ensure compatibility with python 3.7 & 3.8
  • positional only arguments were added in 3.8 - PEP 570
  • the union operator for dictionaries was added in 3.9 - PEP 584
  • defining properties on classmethods was added in 3.9, but removed in 3.11 again
• [x] add searchtype: halton for Active Learning to allow more dimensions
• [x] activate black

  Ideas

• split main into different functions, or better yet: create a Study class
• automatically test all examples
• the integration of the new run system with the Config is now suboptimal. As the defaults are now processed by the components themselves, the config is not filled anymore.
  • however I see no alternative if the components should be accessible with an easy API (without creating a full study/config first)
  • reinsert the processed config into the config class after the components are instantiated? any better ideas?
  • do we even need/want the config class to be filled with all options?

[coveralls]

Pull Request Test Coverage Report for Build 3663821493

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

• For more information on this, see Tracking coverage changes with pull request builds.
• To avoid this issue with future PRs, see these Recommended CI Configurations.
• For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

• 1221 of 1933 (63.17%) changed or added relevant lines in 36 files are covered.
• 345 unchanged lines in 27 files lost coverage.
• Overall coverage increased (+1.2%) to 68.147%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
profit/al/active_learning.py	13	14	92.86%
profit/sur/encoders.py	32	33	96.97%
profit/util/file_handler.py	21	22	95.45%
profit/init.py	0	2	0.0%
profit/run/worker.py	54	56	96.43%
profit/sur/linreg/chaospy_linreg.py	22	24	91.67%
profit/util/component.py	26	28	92.86%
profit/util/util.py	7	10	70.0%
profit/sur/gp/backend/python_kernels.py	9	13	69.23%
profit/run/interface.py	49	54	90.74%
<!--	Total:	1221	1933	63.17%	-->

Files with Coverage Reduction	New Missed Lines	%
profit/run/runner.py	1	86.96%
profit/run/slurm.py	1	15.38%
profit/init.py	2	77.78%
profit/run/worker.py	2	94.19%
profit/util/file_handler.py	2	93.88%
profit/al/simple_al.py	3	83.15%
profit/util/base_class.py	3	84.21%
profit/al/active_learning.py	4	91.53%
profit/sur/gp/backend/python_kernels.py	4	57.14%
profit/sur/gp/gaussian_process.py	4	94.2%
<!--	Total:	345		-->

Totals
Change from base Build 2707154646:	1.2%
Covered Lines:	2854
Relevant Lines:	4188

---

💛 - Coveralls