Crate datarust 

datarust

Scikit-Learn Preprocessing in Rust. A modular, dependency-free data-preprocessing library built on a lightweight Matrix type backed by Vec<Vec<f64>>.

Modules

• scaler — StandardScaler, MinMaxScaler, RobustScaler, MaxAbsScaler, Normalizer, Binarizer, KBinsDiscretizer, QuantileTransformer, PowerTransformer
• encoder — LabelEncoder, OneHotEncoder, OrdinalEncoder, TargetEncoder, FrequencyEncoder
• imputer — SimpleImputer (mean / median / most_frequent / constant) and KnnImputer
• polynomial — PolynomialFeatures
• selection — VarianceThreshold, SelectKBest
• decomposition — PCA, TruncatedSVD
• linear_model — LinearRegression, Ridge, Lasso, LogisticRegression
• metrics — regression metrics (MSE, MAE, R², …) and classification metrics (accuracy, F1, …)
• model_selection — train_test_split, KFold, StratifiedKFold, cross_val_score
• pipeline — sequential Transformer pipelines
• compose — ColumnTransformer
• cluster — KMeans (Lloyd’s algorithm, k-means++ initialization)
• function_transformer — wrap arbitrary functions as a Transformer
• stats — column and 1-D statistics, covariance and correlation matrices
• matrix — Matrix, StrMatrix and SparseMatrix data containers
• serialize — JSON save/load (requires the serde feature)
• transformer_kind — type-erased TransformerKind enum wrapper
• categorical_kind — type-erased CategoricalTransformerKind enum wrapper for encoders
• target_kind — type-erased TargetTransformerKind enum wrapper for supervised encoders

All numeric transformers implement the Transformer trait. Supervised estimators implement Predictor (fit with features + target, then predict); regressors additionally implement Regressor and classifiers implement Classifier / PredictProba where appropriate. Categorical encoders (OneHot, Ordinal, Frequency) implement the CategoricalTransformer trait. The TargetEncoder implements the TargetTransformer trait (requires target values during fit). The LabelEncoder implements the LabelTransformer trait (1-D string ↔ int mapping). Clustering estimators (KMeans) implement the Clusterer trait (fit on X only, then predict returning cluster indices).

Features

• serde — enables JSON serialization via serialize.
• rayon — enables parallel column/row operations for large datasets.
• matrixmultiply — enables a tuned pure-Rust GEMM (no system BLAS) for Matrix::matmul and covariance computation, speeding up PCA and TruncatedSVD on large dense inputs.
• datasets — embeds classic toy datasets (Iris, Breast Cancer, Wine, Diabetes) as const arrays for examples, tests, and onboarding.

The default build has zero external dependencies.

Re-exports

pub use categorical_kind::CategoricalTransformerKind;

pub use cluster::KMeans;

pub use cluster::KMeansInit;

pub use compose::ColumnSpec;

pub use compose::ColumnTransformer;

pub use compose::Output;

pub use compose::Remainder;

pub use compose::Table;

pub use encoder::DropStrategy;

pub use encoder::FrequencyEncoder;

pub use encoder::HandleUnknown;

pub use encoder::LabelEncoder;

pub use encoder::OneHotEncoder;

pub use encoder::OrdinalCategories;

pub use encoder::OrdinalEncoder;

pub use encoder::OrdinalHandleUnknown;

pub use encoder::TargetEncoder;

pub use encoder::UnknownFrequency;

pub use encoder::UnknownTarget;

pub use error::DatarustError;

pub use error::Result;

pub use label_space::LabelSpace;

pub use linear_model::Lasso;

pub use linear_model::LinearRegression;

pub use linear_model::LinearSolver;

pub use linear_model::LogisticRegression;

pub use linear_model::LogisticSolver;

pub use linear_model::Ridge;

pub use linear_model::RidgeSolver;

pub use matrix::Matrix;

pub use matrix::SparseMatrix;

pub use matrix::StrMatrix;

pub use model_selection::cross_val_score;

pub use model_selection::train_test_split;

pub use model_selection::KFold;

pub use model_selection::StratifiedKFold;

pub use model_selection::TrainTestSplit;

pub use pipeline::Pipeline;

pub use pipeline::SupervisedPipeline;

pub use target_kind::TargetTransformerKind;

pub use traits::default_input_names;

pub use traits::CategoricalTransformer;

pub use traits::Classifier;

pub use traits::Clusterer;

pub use traits::Estimator;

pub use traits::FeatureNames;

pub use traits::LabelTransformer;

pub use traits::ParamValue;

pub use traits::Params;

pub use traits::PredictProba;

pub use traits::Predictor;

pub use traits::Regressor;

pub use traits::TargetTransformer;

pub use traits::Transformer;

pub use transformer_kind::TransformerKind;

Modules

categorical_kind

Enum wrapper for categorical transformer types, enabling serialization of ColumnTransformer under the serde feature.

cluster

Unsupervised clustering estimators.

compose

Composing multiple transformers.

datasets

Classic toy datasets for examples, tests, and onboarding.

decomposition

Dimensionality reduction: PCA and Truncated SVD.

encoder

Categorical encoders.

error

Error types returned by fallible operations.

function_transformer

Wrap arbitrary functions as a Transformer.

imputer

Missing-value imputation.

label_space

Compact mapping between original class labels and dense model indices. Compact vocabulary for non-negative integer-valued class labels.

linalg

Shared linear-algebra primitives (Cholesky solver, etc.). Linear-algebra primitives shared by estimators.

linear_model

Regression & classification estimators: LinearRegression, Ridge, Lasso, LogisticRegression. Regression estimators with a fit/predict API.

matrix

Dense and sparse matrix containers used throughout the crate.

metrics

Model-evaluation metrics: regression (MSE, R², …) and classification (accuracy, F1, …). Model-evaluation metrics mirroring sklearn.metrics.

model_selection

Model selection: train_test_split, KFold, cross_val_score. Model-selection utilities mirroring sklearn.model_selection.

pipeline

Sequential transformer pipelines.

polynomial

Generate polynomial feature combinations.

scaler

Feature scaling and discretization transformers.

selection

Feature selection.

serialize

Optional JSON serialization for fitted transformers (gated on the serde feature).

stats

Column-wise statistics, covariance and correlation helpers.

target_kind

Enum wrapper for target (supervised) categorical transformers, enabling serialization of ColumnTransformer under the serde feature.

traits

Core traits shared by the numeric and categorical transformers.

transformer_kind

Enum wrapper for all transformer types, enabling serialization of Pipeline and ColumnTransformer under the serde feature.