Improve argument processing of tranquilo

[janosg]

Problems

1. The default configuration of tranquilo is quite complex. This is easiest shown with an example: The default target_sample_size depends on the model_type; The default model_type depends on the functype. But the model_type can also be overwritten by the user. If so, we need to check that the user specified a valid model type before we can create the default target_sample_size. I.e. everything is entangled and the order of processing stuff has to be chosen very carefully.
2. The default values are not collected in one place.
3. The namespace of the main tranquilo function is cluttered with stuff that was only needed to initialize the main components; It is hard to tell which variables will get used later on and which can be easily changed.

Solution

• All default values should be defined in options.py; If default values depend on other arguments, we implement getter_functions, e.g. get_default_radius_options(x)
• Nothing else is in options.py; In particular, all the logic related to combining default options and user options is in process_arguments.py.
• The outer tranquilo function only calls process_arguments and then calls _internal_tranquilo with the processed arguments. This removes all namespace cluttering. A small trick to avoid long lists of arguments, the outer tranquilo functions only uses *args, **kwargs as arguments and functools.wraps(process_arguments) to get the right signature and docstring.

Alternatives

We could use dags to figure out the order of everything, but we currently do not have dags as a dependency and we don't need it anywhere else. In any case, the current PR gets us closer to a situation where we could use dags.

To-Do

• [x] Move the actual logic from tranquilo into _internal_tranquilo
• [x] Rename _new_tranquilo to _tranquilo
• [ ] Run all tests and a benchmark
• [x] Make RadiusOptions.initial_radius a mandatory argument without default value
• [x] Make all ConvOptions mandatory
• [x] Make all StopOptions mandatory
• [x] Write some tests for the functions in process_arguments and options.py
• [x] Put all component default options into options.py. Example:
  • [x] Define a class SubsolverOptions that contains everything the default_options dict in solve_subproblem contains
  • [x] Adjust get_component sucht that the user_options can be NamedTuples on which we call _asdict
  • [x] Instantiate the new class in solve_subproblem instead of defining a dict there
• [x] Improve error message in default aggregator and move the entire compatibility check from aggregate_models into get_default_aggregator if possible.
• [x] make default model fitter dependent on model type and sample size instead of functype
• [ ] Improve error handling and type checking inside process_arguments

Discuss

• Check up on differences in benchmarks introduced by commit 9d68fab
• Double-check if we want to divide AcceptanceOptions into individual options in acceptance_decision.py.

[codecov[bot]]

Codecov Report

Merging #446 (bf07d2a) into main (91ffcb8) will increase coverage by 0.05%. The diff coverage is 99.69%.

@@            Coverage Diff             @@
##             main     #446      +/-   ##
==========================================
+ Coverage   92.99%   93.05%   +0.05%     
==========================================
  Files         247      250       +3     
  Lines       18482    18597     +115     
==========================================
+ Hits        17188    17305     +117     
+ Misses       1294     1292       -2     

Impacted Files	Coverage Δ
...sts/optimization/subsolvers/test_gqtpar_lambdas.py	100.00% <ø> (ø)
...optimization/tranquilo/test_acceptance_decision.py	100.00% <ø> (ø)
tests/optimization/tranquilo/test_rho_noise.py	100.00% <ø> (ø)
src/estimagic/optimization/tranquilo/options.py	99.18% <98.92%> (-0.82%)	:arrow_down:
...agic/optimization/tranquilo/acceptance_decision.py	96.87% <100.00%> (ø)
...c/optimization/tranquilo/acceptance_sample_size.py	95.45% <100.00%> (ø)
...timagic/optimization/tranquilo/aggregate_models.py	97.67% <100.00%> (+10.17%)	:arrow_up:
...imagic/optimization/tranquilo/estimate_variance.py	100.00% <100.00%> (ø)
.../estimagic/optimization/tranquilo/filter_points.py	79.83% <100.00%> (+0.16%)	:arrow_up:
src/estimagic/optimization/tranquilo/fit_models.py	84.51% <100.00%> (-1.85%)	:arrow_down:
... and 10 more

... and 4 files with indirect coverage changes

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.