results_cluster_tools.py

Custom Cluster Test — exhaustive pipeline search against known components.

---

Constants

Name	Value
EXTERNAL_METRICS	{'ARI': {'display': 'Adjusted Rand Index', 'func': lambda…
DEFAULT_EXTERNAL_METRICS	['ARI', 'AMI', 'V-measure']
PRIMARY_EXTERNAL_METRIC	'ARI'
OTHER_LABEL_NAME	'other'
ALGO_PARAM_SPECS	{'K-Means': {'density': False, 'needs_k': True, 'params':…
ALGORITHMS	list(ALGO_PARAM_SPECS.keys())
DATA_TYPES	list(DATA_KEY_MAP.keys())
SCALINGS	['None', 'Robust Z-score', 'CLR', 'ILR']
DIM_REDUCTIONS	['None', 'PCA', 't-SNE']
DEFAULT_DATA_TYPES	['Counts']
DEFAULT_SCALINGS	['None']
DEFAULT_DIM_REDUCTIONS	['None']
DEFAULT_ALGORITHMS	['K-Means']

Classes

Preprocessor

Builds and caches preprocessed matrices for the sweep.

Method	Signature	Description
__init__	(self, particle_data, elements, filter_zeros=True, tsne_random_state=4	Initialise the cache and the fixed kept-row mask.
n_rows	(self)	Number of kept particles.
raw_matrix	(self, data_type)	Return the kept, unscaled matrix for data_type (cached).
counts_matrix	(self)	Return the kept count matrix used to build ground truth.
_scaled	(self, data_type, scaling)	Return the scaled matrix for one (data_type, scaling) (cached).
matrix	(self, data_type, scaling, dim_reduction)	Return the fully preprocessed matrix for one pipeline (cached).

Functions

Function	Signature	Description
_v_measure	(truth, pred)	V-measure (harmonic mean of homogeneity and completeness).
_homogeneity	(truth, pred)	Homogeneity score: each cluster contains a single truth class.
_completeness	(truth, pred)	Completeness score: each truth class falls into a single cluster.
multiplicative_replacement	(matrix, frac=0.65, threshold=None)	Replace zeros in a non-negative composition matrix without distorting ratios.
parse_components	(text)	Parse a component string into [(name, [elements]), ...].
build_ground_truth	(raw_matrix, elements, components, other_flags=None)	Assign each particle to a named component or to "other".
_row_for_particle	(p, data_type, elements)	Build one matrix row for a particle in the requested representation.
run_algorithm	(name, params, data, som_runner=None)	Fit one algorithm with explicit params and return integer labels.
make_host_som_runner	(host_dialog)	Build a SOM runner that does not disturb the host's SOM tab state.
build_param_grid	(name, selections)	Expand per-parameter value lists into concrete parameter dicts.
count_combinations	(pre_combos, algo_selections)	Return the total number of fits a sweep would perform.
_params_str	(algo, params)	Return a compact human-readable parameter string for a result row.
run_sweep	(particle_data, elements, components, *, data_types, scalings, dim_red	Run the full pipeline grid and score every result against ground truth.
rank_results	(results, metric=PRIMARY_EXTERNAL_METRIC)	Return results sorted best-first by metric (NaNs last).
_spearman	(a, b)	Return the tie-corrected Spearman rank correlation of two sequences.
borda_count_rank	(results, metrics, registry=None)	Rank results by Borda count across metrics (internal or external).
analyze_metric_trust	(results, internal_metrics, reference=PRIMARY_EXTERNAL_METRIC)	Report how well each internal index tracks the ground-truth reference.
analyze_metric_trust_stratified	(results, internal_metrics, reference=PRIMARY_EXTERNAL_METRIC, min_per	Validate each internal index against ground truth within fixed preprocessing.
summarize_sweep_failures	(failures, total=None)	Summarise where a sweep produced no usable partition, per algorithm.
per_cluster_silhouette	(data, labels)	Return the mean silhouette width of each individual cluster.