Algorithms

The catalog mirrors the complete scikit-learn taxonomy. Every item enters the roadmap with a mandatory four-stage evolution — naive → tests → performance → out-of-core — and receives a Python binding as soon as it is validated on CPU.

39+ algorithms

Preprocessing5
Imputation2
Linear models7
Nearest neighbors4
SVM3
Trees2
Ensembles3
Clustering3
Decomposition2
Outliers1+
Model selection3
Metrics4+

Preprocessing · `sciencekit_preprocessing`

Estimator	Description	Status
`SKStandardScaler`	Standardizes by removing the mean and scaling to unit variance	Planned
`SKMinMaxScaler`	Scales each feature to a fixed range	Planned
`SKRobustScaler`	Scaling robust to outliers through quartile statistics	Planned
`SKOneHotEncoder`	One-hot encoding of categories	Planned
`SKPolynomialFeatures`	Polynomial and interaction features	Planned

Imputation · `sciencekit_impute`

Estimator	Description	Status
Simple strategies	Mean, median, most frequent and constant	Planned
`SKKNNImputer`	Multivariate imputation via nearest neighbors	Planned

Linear models · `sciencekit_linear_model`

Estimator	Description	Status
`SKLinearRegression`	Ordinary least squares	Planned
`SKRidge`	Regression with L2 regularization	Planned
`SKLasso`	Regression with L1 regularization (sparse support)	Planned
`SKElasticNet`	Combined L1 + L2 regularization	Planned
`SKLogisticRegression`	Regularized logistic classification	Planned
`SKSGDClassifier`	Incremental classification via SGD · streaming (`SKLazySource`)	Planned
`SKSGDRegressor`	Incremental regression via SGD · streaming (`SKLazySource`)	Planned

Nearest neighbors · `sciencekit_neighbors`

Structure/Estimator	Description	Status
`SKKNeighborsClassifier`	Classification by k nearest neighbors	Planned
`SKKNeighborsRegressor`	Regression by k nearest neighbors	Planned
`SKKDTree`	kd-tree for fast spatial search	Planned
`SKBallTree`	Ball tree for arbitrary metrics	Planned

Support vector machines · `sciencekit_svm`

Estimator	Description	Status
`SKSVC`	Kernel-based support vector classification	Planned
`SKSVR`	Support vector regression	Planned
`SKLinearSVC`	Scalable linear SVM (dense and sparse)	Planned

Trees and ensembles · `sciencekit_tree` + `sciencekit_ensemble`

Estimator	Description	Status
`SKDecisionTreeClassifier`	Decision tree for classification	Planned
`SKDecisionTreeRegressor`	Decision tree for regression	Planned
`SKRandomForest`	Random forest aggregated in parallel with rayon	Planned
`SKBagging`	Parallel bootstrap aggregation	Planned
`SKGradientBoosting`	Sequential gradient boosting	Planned

Unsupervised · `sciencekit_cluster`, `sciencekit_decomposition`, `sciencekit_outlier`

Estimator	Description	Status
`SKKMeans`	K-means with O(1) random access via memmap	Planned
`SKMiniBatchKMeans`	Mini-batch k-means with streaming (`SKLazySource`)	Planned
`SKDBSCAN`	Density-based clustering	Planned
`SKPCA`	Principal component analysis	Planned
`SKTruncatedSVD`	Truncated SVD for sparse data	Planned
Anomaly detection	Outlier estimators from the scikit-learn family	Planned

Model selection, pipelines and metrics

Item	Crate	Description	Status
`sk_train_test_split`	`sciencekit_model_selection`	Deterministic data partitioning	Planned
`SKKFold`	`sciencekit_model_selection`	k-fold cross-validation	Planned
`SKGridSearchCV`	`sciencekit_model_selection`	Grid search with cross-validation	Planned
`SKPipeline`	`sciencekit_pipeline`	Type-safe pipeline (associated types) and DAGs	Planned
Full metrics	`sciencekit_metrics`	accuracy, f1, MSE, confusion matrix and the remaining sklearn metrics	Planned

Continuous acceleration: every completed algorithm receives, in separate changes, a Python binding (sciencekit_python) and GPU backends (OpenCL → CUDA → ROCm) behind SKComputeBackend. The CPU is the default backend, always present.

Next: architecture → I want to implement one of these ⭐

Keyboard shortcuts

sciencekit