Experiments

The notebook workflows live here. Each experiment folder keeps its plots, exported CSV files, sweep results, and notebook variants together.

Folders

1. model_predict_cavity_claw_RouteMeander_eigenmode handles resonator and cavity claw route meander eigenmode prediction.
2. model_predict_coupler_NCap_cap_matrix handles coupler NCap capacitance matrix prediction.
3. model_predict_qubit_TransmonCross_cap_matrix handles transmon cross capacitance matrix prediction.
4. model_predict_qubit_TransmonCross_Hamiltonian_params handles transmon cross Hamiltonian parameter prediction and stress test analysis.

Common Conventions

1. ml_00_* contains data loading and exploratory analysis.
2. ml_01_* contains primary training notebooks.
3. ml_02_* contains result inspection and CSV export.
4. ml_03_* contains hyperparameter analysis.
5. ml_10_* and ml_20_* contain surrogate model or alternative loss variants.
6. validation_* contains downstream simulation and EM simulator validation notebooks.
7. parameters*.py contains notebook configuration for each training variant.

model_predict_qubit_TransmonCross_Hamiltonian_params is the furthest developed experiment and extends past this range. It starts at ml_10, because data preparation is shared with the capacitance matrix experiment, and adds the stress test (ml_14), the runtime benchmarks (ml_22, ml_30, ml_31), and the sweep and multi seed studies (ml_32 through ml_40).

Inverse plus surrogate

The Hamiltonian parameter experiment trains in both directions. A forward surrogate maps cross claw geometry to (f_q, alpha). An inverse model then maps a requested (f_q, alpha) back to geometry, trained with that surrogate frozen so the loss is evaluated in Hamiltonian space and only the inverse weights update. Predicted geometries are validated with a conventional EM solver in the loop.

Large local artifacts such as cached datasets, trained models, and scalers are ignored through the root .gitignore.

Each experiment root keeps notebooks at top level while ancillary files are grouped into focused folders.

1. metadata contains schema files, saved column lists, and design or config JSON.
2. results/training contains loss histories and test summaries.
3. results/predictions contains exported prediction and reconstruction CSV files.
4. results/validation contains sweep summaries, candidate exports, and EM simulator comparison tables.