Python API reference¶

All names below are exported from squeeze. Constructor signatures are recorded from the built package and checked by the documentation smoke command. This is an inventory of the current interface, not a promise of sklearn compatibility.

Core reducers¶

Rust methods consume dense float64 matrices. Except PCA, they primarily return a new embedding from fit_transform(X). See the algorithm capability table.

UMAP¶

squeeze.UMAP(n_neighbors=15, n_components=2, metric='euclidean', metric_kwds=None, output_metric='euclidean', output_metric_kwds=None, n_epochs=None, learning_rate=1.0, init='spectral', min_dist=0.1, spread=1.0, low_memory=True, n_jobs=-1, set_op_mix_ratio=1.0, local_connectivity=1.0, repulsion_strength=1.0, negative_sample_rate=5, transform_queue_size=4.0, a=None, b=None, random_state=None, angular_rp_forest=False, target_n_neighbors=-1, target_metric='categorical', target_metric_kwds=None, target_weight=0.5, transform_seed=42, transform_mode='embedding', force_approximation_algorithm=False, verbose=False, tqdm_kwds=None, unique=False, densmap=False, dens_lambda=2.0, dens_frac=0.3, dens_var_shift=0.1, output_dens=False, disconnection_distance=None, precomputed_knn=(None, None, None), use_hnsw=None, hnsw_prune_strategy='simple', hnsw_alpha=1.2) -> None

Usage, parameters and limitations.

PCA¶

squeeze.PCA(n_components=2)

Usage, parameters and limitations.

TSNE¶

squeeze.TSNE(n_components=2, perplexity=30.0, learning_rate=200.0, n_iter=1000, early_exaggeration=12.0, random_state=None, theta=0.5, use_barnes_hut=None, min_grad_norm=1e-07, n_iter_without_progress=300)

Usage, parameters and limitations.

MDS¶

squeeze.MDS(n_components=2, metric=True, n_iter=300, random_state=None)

Usage, parameters and limitations.

Isomap¶

squeeze.Isomap(n_components=2, n_neighbors=10)

Usage, parameters and limitations.

LLE¶

squeeze.LLE(n_components=2, n_neighbors=12, reg=0.001, error_on_singular=False)

Usage, parameters and limitations.

PHATE¶

squeeze.PHATE(n_components=2, k=15, t=5, decay=2.0, random_state=None)

Usage, parameters and limitations.

TriMap¶

squeeze.TriMap(n_components=2, n_inliers=12, n_outliers=4, n_random=3, n_iter=800, learning_rate=0.1, weight_adj=50.0, random_state=None)

Usage, parameters and limitations.

PaCMAP¶

squeeze.PaCMAP(n_components=2, n_neighbors=10, mn_ratio=0.5, fp_ratio=2.0, n_iter=450, learning_rate=1.0, random_state=None)

Usage, parameters and limitations.

NeighborMap¶

squeeze.NeighborMap(n_neighbors=15, n_epochs=160, negative_samples=5, learning_rate=1.0, init='pca', random_state=42)

Usage, parameters and limitations.

SpectralMap¶

squeeze.SpectralMap(n_neighbors=15, n_iter=128, random_state=42)

Usage, parameters and limitations.

UMAP-family interfaces¶

AlignedUMAP fits related datasets with explicit row relations. ParametricUMAP requires its optional TensorFlow training stack; its dependency-error placeholder is not the actual model signature. See UMAP variants.

Composition and extensions¶

DRPipeline¶

squeeze.DRPipeline(steps: 'list[tuple[str, Any]]') -> 'None'

EnsembleDR¶

squeeze.EnsembleDR(methods: 'list[tuple[str, Any, float]]', blend_mode: 'str' = 'weighted_average', alignment: 'str' = 'procrustes') -> 'None'

ProgressiveDR¶

squeeze.ProgressiveDR(coarse: 'Any', fine: 'Any', blend_steps: 'int' = 10, blend_function: 'str' = 'linear') -> 'None'

AdaptiveDR¶

squeeze.AdaptiveDR(method_map: 'dict[str, Any]', strategy: 'str' = 'size') -> 'None'

OutOfSampleDR¶

squeeze.OutOfSampleDR(base_reducer: 'Any', n_neighbors: 'int' = 5, weights: 'str' = 'distance') -> 'None'

StreamingDR¶

squeeze.StreamingDR(base_reducer: 'Any', n_neighbors: 'int' = 5) -> 'None'

Read composition, transforms, and streaming for fit/refit behavior and inference limitations.

Evaluation¶

DREvaluator evaluates a pair of original/reduced arrays. EvaluationReport stores its outputs and exposes summary() and to_dict(). Overlap-based legacy names are distinct from sklearn rank trustworthiness; see metric definitions.

DREvaluator¶

Evaluation entry point; expensive stability checks can be disabled.

squeeze.DREvaluator(X_original: 'np.ndarray', X_reduced: 'np.ndarray', labels: 'np.ndarray | None' = None, reducer: 'BaseEstimator | None' = None, method_name: 'str' = 'Unknown') -> 'None'

quick_evaluate¶

Returns overlap-based trustworthiness/continuity plus Spearman correlation.

squeeze.quick_evaluate(X_original: 'np.ndarray', X_reduced: 'np.ndarray', k: 'int' = 15) -> 'dict[str, float]'

trustworthiness¶

Legacy name for neighbor-overlap quality at k; not a rank penalty.

squeeze.trustworthiness(X_original: 'np.ndarray', X_reduced: 'np.ndarray', k: 'int' = 15) -> 'float'

continuity¶

Legacy overlap-based continuity calculation at k.

squeeze.continuity(X_original: 'np.ndarray', X_reduced: 'np.ndarray', k: 'int' = 15) -> 'float'

co_ranking_quality¶

Overlap-based neighborhood quality, not a full co-ranking matrix.

squeeze.co_ranking_quality(X_original: 'np.ndarray', X_reduced: 'np.ndarray', k: 'int' = 15) -> 'float'

spearman_distance_correlation¶

Correlates pairwise distances; optionally caps sample count.

squeeze.spearman_distance_correlation(X_original: 'np.ndarray', X_reduced: 'np.ndarray', max_samples: 'int | None' = 5000) -> 'float'

global_structure_preservation¶

Compares distances between class centroids; requires labels.

squeeze.global_structure_preservation(X_original: 'np.ndarray', X_reduced: 'np.ndarray', labels: 'np.ndarray') -> 'float'

local_density_preservation¶

Compares local density estimates at k.

squeeze.local_density_preservation(X_original: 'np.ndarray', X_reduced: 'np.ndarray', k: 'int' = 15) -> 'float'

reconstruction_error¶

Returns reconstruction error measures; only the linear method is implemented.

squeeze.reconstruction_error(X_original: 'np.ndarray', X_reduced: 'np.ndarray', method: 'str' = 'linear') -> 'dict[str, float]'

clustering_quality¶

Scores clustering in the embedding; labels enable agreement metrics.

squeeze.clustering_quality(X_reduced: 'np.ndarray', labels_true: 'np.ndarray | None' = None, n_clusters: 'int | None' = None) -> 'dict[str, float]'

classification_accuracy¶

Cross-validates a classifier on the supplied embedding; it does not refit the embedding per fold.

squeeze.classification_accuracy(X_reduced: 'np.ndarray', labels: 'np.ndarray', cv: 'int' = 5, classifier: 'BaseEstimator | None' = None) -> 'dict[str, float]'

bootstrap_stability¶

Refits on resampled data and compares stability; additional fit cost.

squeeze.bootstrap_stability(X: 'np.ndarray', reducer: 'BaseEstimator', n_bootstrap: 'int' = 10, sample_fraction: 'float' = 0.8, random_state: 'int | None' = None) -> 'dict[str, float]'

noise_robustness¶

Refits with noise at each supplied level.

squeeze.noise_robustness(X: 'np.ndarray', reducer: 'BaseEstimator', noise_levels: 'list[float] | None' = None, random_state: 'int | None' = None) -> 'dict[float, float]'

parameter_sensitivity¶

Refits a reducer class across a parameter grid.

squeeze.parameter_sensitivity(X: 'np.ndarray', reducer_class: 'type', parameters_to_vary: 'dict[str, list[Any]] | None' = None, base_params: 'dict[str, Any] | None' = None, random_state: 'int | None' = None) -> 'dict[str, dict[str, Any]]'

Strategy registry¶

STRATEGIES is the shared instance of StrategyRegistry. Strategy stores a name, algorithm class, default parameters, description and category. StrategyRegistry supplies register, get, create, names, categories, by_category, summary, and iteration. See registry usage.

squeeze.get_strategy(name: 'str') -> 'Strategy'

squeeze.list_strategies() -> 'list[str]'

squeeze.create_reducer(name: 'str', **kwargs)

Version and source¶

squeeze.__version__ reports installed package metadata, falling back to a development version when metadata is unavailable. Record the Git commit as well for experiments.