[perrymok]

I have asked Gemini to create some tests. They look reasonable.

Added Test Modules in tests

1. test_base.py
   • test_eelsnmf_basic: EELSNMF object initialization, G matrix building, and default NMF decomposition.
   • test_eelsnmf_metadata_init: Auto-extraction of beam parameters ($E_0,\alpha,\beta$) from HyperSpy signal metadata.
   • test_change_dtype: Changing internal array dtypes (e.g. float32).
   • test_temp_arrays: Dynamic array allocation and cleanup (create_temp_array, delete_temp_arrays).
   • test_save_and_load: Pickling/unpickling model objects using save() and load().
2. test_utils.py
   • Array & Grid utilities: norm(), find_2factors(), find_index(), convolve(), moving_average(), all_arrays_equal(), and match_axis().
   • Physics helpers: $\theta_E$, convergent_factor(), convergent_psi(), integral_over_th(), psi().

• ListOfSI dataset aggregation: Spatial unfolding/folding and save/load_ListOfSI.

3. test_modelG.py
   • Delta model G matrix construction with Kohl and Zezhong cross-sections.
   • Background exponent specifications (int, array, None, invalid types).
   • available_models property checks.
4. test_decompositions.py
   • NMF initialization algorithms (nndsvd, random with random seed).
   • Fixed matrix constraints (W_fixed_bool, W_fixed_values).
   • Custom initial matrices (W_init, H_init).
   • KL-divergence decomposition (default_kl_decomposition).
   • Matrix reconstruction (get_model()) and enforce_dtype().
5. test_analysis.py
   • Spatial loading shape computation (calculate_loadings()).
   • Component ELNES spectrum extraction (get_edge_from_component()).
   • Chemical quantification (quantify_components(), spatial_standard_q(), get_chemical_maps()).
   • Figures of merit evaluation (evaluate_comp_contributions() for leave-out loss, power, and explained variance).
6. test_plot.py
   • Non-interactive rendering verification for all plot utilities (plot_factors, plot_loadings, plot_edges, plot_chemical_maps, plot_average_model, plot_energy_ranges).