refactor: replace search YAML config with explicit Python defaults

[Jammy2211]

Overview

Search classes (Nautilus, Dynesty, Emcee, Zeus, BFGS, Drawer) load default parameters from YAML config files via a blind setattr() loop with no type safety. This makes the source code hard to follow, new samplers hard to add, and config typos hard to detect. Workspace scripts already pass all parameters explicitly, so the YAML defaults are unused in practice.

This refactor moves all search/run parameter defaults into explicit typed __init__ arguments, eliminates config_dict_search/config_dict_run entirely in favour of direct instance attributes, and removes the per-search YAML files.

Plan

• Move search and run parameter defaults into explicit __init__ arguments on each search class with type annotations
• Eliminate config_dict_search and config_dict_run — replace all dict access with direct instance attributes
• Remove config machinery from the base class (setattr() loop, config_type, _class_config, _config())
• Delete per-search YAML files (nest.yaml, mcmc.yaml, mle.yaml); keep general.yaml for cross-cutting settings
• Update tests to assert against instance attributes instead of config-loaded dict values

Detailed implementation plan

Affected Repositories

• PyAutoFit (primary)

Work Classification

Library

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoFit	main	clean

Suggested branch: feature/search-config-cleanup
Worktree root: ~/Code/PyAutoLabs-wt/search-config-cleanup/

Implementation Steps

1. Refactor NonLinearSearch.__init__ — remove setattr() loop, config_dict properties, config_type, _class_config, _config(). Add silence as explicit param. Keep general.yaml reads.
2. Refactor Nautilus — explicit typed params in __init__, replace all config_dict_search["key"] with self.key, replace **config_dict_search with explicit kwargs to sampler.
3. Refactor DynestyStatic/DynestyDynamic/AbstractDynesty — same pattern.
4. Refactor Emcee/Zeus/AbstractMCMC — same pattern, keep AutoCorrelationsSettings class but with Python defaults.
5. Refactor BFGS/LBFGS/Drawer/AbstractMLE — same pattern, options become explicit params.
6. Remove AbstractNest.config_type and related overrides.
7. Delete autofit/config/non_linear/{nest,mcmc,mle}.yaml.
8. Update tests to assert instance attributes directly.
9. Update mock search to remove config_dict_search dependency.

Key Files

• autofit/non_linear/search/abstract_search.py — base class config machinery removal
• autofit/non_linear/search/nest/nautilus/search.py — Nautilus explicit params
• autofit/non_linear/search/nest/dynesty/search/{static,dynamic,abstract}.py — Dynesty explicit params
• autofit/non_linear/search/mcmc/{emcee,zeus}/search.py — MCMC explicit params
• autofit/non_linear/search/mle/{bfgs,drawer}/search.py — MLE explicit params
• autofit/config/non_linear/*.yaml — delete

Original Prompt

Click to expand starting prompt

All searches use config files to load their input parameters and interface with their respective source code library.

Part of the design was so that a project like PyAutoLens could create a search, have the parameters set in a way that was suitable for that model and hide from the user everything else:

search = af.Nautilus(
path_prefix=Path("imaging"),
name="start_here",
unique_tag=dataset_name,
n_live=100,
n_batch=50,
iterations_per_quick_update=1000,
)

Example of a modeling script:

@autolens_workspace/scripts/imaging/modeling.py

However, the downside of this approach is the autofit source code got very confusing with search config loads, the config files which control this for a project are bloated and require energy to set up and maintain and it inevitably leads to bug.

This issue shows part of the problem:

#1001

Is there a solution which makes the source code interface a lot cleaner, doesnt have these config files and makes it easier for new samplers to be set up? I wonder if having a Python class input to each search class is the solution, which in the past would be a pain to maintain for each specific sampler but now AI agents can do this quickly isnt so bad.

The biggest downside is a project like autolens_workspace may need to manually specify search inputs everywhere the search is set up, albeit maybe theres a no-required-config trick or inheritance trick we could use?

Think hard and come up with an assessment of how we can make the source code a lot cleaner without losing the desired API.

[Jammy2211]

Completed. All search parameters now have explicit Python defaults — YAML config files removed. See PR #1202.