photonics2d.optimize() raises KeyError: 'num_optimization_steps' after Config refactor

[mkeeler43]

Bug

After commit 0c6f17c (refactor(photonics): make num_optimization_steps an attribute of Config), calling optimize() on the photonics2d problem raises a KeyError when num_optimization_steps is not explicitly included in the config dict.

KeyError: 'num_optimization_steps'

Cause

The commit made two changes:

1. Moved num_optimization_steps from a class-level default (_num_optimization_steps_default = 200) into the Config dataclass (which extends Conditions)
2. Changed the access pattern from a safe .get() with fallback to a hard key lookup:

- num_optimization_steps = conditions.get("num_optimization_steps", self._num_optimization_steps_default)
+ num_optimization_steps = conditions["num_optimization_steps"]

However, _setup_simulation merges the passed config with self.conditions, which is the upcast Conditions instance (via upcast(self.config)). Since num_optimization_steps is a Config field but not a Conditions field, it gets stripped during the upcast and is never present in the merged dict — unless the caller explicitly passes it.

# v0.py:176 — self.conditions has no num_optimization_steps
current_conditions = {**dataclasses.asdict(self.conditions), **(config or {})}

Reproduction

Any call to problem.optimize() where the config dict doesn't explicitly include num_optimization_steps will fail:

from engibench.utils.all_problems import BUILTIN_PROBLEMS

problem = BUILTIN_PROBLEMS["photonics2d"]()
problem.reset(seed=1)
design = problem.dataset["test"]["optimal_design"][0]
problem.optimize(design, config={"wavelength1": 1.0, "wavelength2": 1.3, "blur_radius": 2})
# KeyError: 'num_optimization_steps'

Suggested Fix

Change _setup_simulation (line 176) to merge against self.config instead of self.conditions, so that Config-level defaults (including num_optimization_steps) are included:

- current_conditions = {**dataclasses.asdict(self.conditions), **(config or {})}
+ current_conditions = {**dataclasses.asdict(self.config), **(config or {})}

[linear]

SOH-12

[g-braeunlich]

Sounds reasonable to me

[markfuge]

Agreed @mkeeler43 feel free to move forward with a PR on this

[g-braeunlich]

I think @mkeeler43 assigned me. Also would be happy to do it. Not a big deal.

[markfuge]

Ah great yes @g-braeunlich take it away!