CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.6.14]

Added

• decomposition/lda.rs: LDA, linear discriminant analysis for supervised dimensionality reduction (#136). It projects the data onto the directions that best separate the classes, keeping min(n_classes - 1, n_features) components by default, and implements the Transformer interface next to PCA. Directions match scikit-learn's LinearDiscriminantAnalysis(solver="eigen") up to sign.

[0.6.13]

Fixed

• xgboost/xgb_regressor.rs: XGRegressor::fit no longer panics when subsample is less than 1.0 on a small dataset (#444). The sample for each tree now keeps a minimum of one row, as scikit-learn does for its own subsample parameter. Sample sizes of one row or more are unchanged.

[0.6.11]

Fixed

• algorithm/neighbour/cosinepair.rs: CosinePair::query_row_top_k now returns exact nearest neighbours whenever approximate is false (the default). Previously the query always sampled only top_k evenly strided candidate rows without documentation, and the bounded candidate heap evicted its closest entry, so the method could return the farthest of the sampled rows (#442). Strided sampling is now gated behind CosinePairParameters { approximate: true, .. } and is documented as approximate.
• algorithm/neighbour/cosinepair.rs: CosinePair construction now evaluates each unordered row pair once (symmetric half-scan), precomputes row norms once in O(n·d), and scores pairs through zero-copy row views instead of materialising two Vecs per pair (#442). Distances are unchanged (bit-identical formula and operation order as Cosine::new().distance(...)); measured build time drops ~3x on a 1500x64 input. Construction remains Theta(n^2) dot products — top_k does not make it sub-quadratic; module and method docs now state both facts.

Changed

• Breaking: CosinePair gained a private row_norms field holding the precomputed row norms. Construct the structure through new / with_top_k / with_parameters instead of struct literals.

[0.6.10]

Fixed

• model_selection: pinned the KFold seed in the test_cross_val_predict_knn and test_cross_validate_knn unit tests. Under --all-features (std_rand) an unseeded KFold draws OS entropy, so each CI run shuffled the folds differently; a sweep of 20 000 seeds showed 0.17% of shuffles violate the MAE < 10.0 assertion (worst 12.81) and 0.01% violate train_score < test_score, making CI flaky. Library behaviour is unchanged; the entropy-seeded path stays covered by the rand_custom tests.

[0.6.9]

Fixed

• metrics: precision, recall, and f1 (the free functions, the Precision / Recall / F1 metric structs, and the matching ClassificationMetrics entry points) now accept any label type that implements Number, including ordered integers such as u16 or i32; labels no longer need to implement RealNumber or FloatNumber (#322). The same integer labels can now feed RandomForestClassifier::fit and classification metrics inside model_selection::cross_validate. Class keys are derived through a shared f64 conversion instead of raw float bit transmutation; scores for float inputs are unchanged.

[0.6.8]

Fixed

• linear/linear_regression.rs: LinearRegression::fit / fit_matrix now return Err(Failed::fit(...)) instead of panicking when the intercept-augmented system is underdetermined, i.e. n_features + 1 > n_samples (#435). Both the default SVD solver and the QR solver are covered.
• linalg/traits/svd.rs: svd_solve / svd_solve_mut reject systems where A has more columns than rows with FailedError::SolutionFailed instead of writing the solution out of b's bounds; SVD::solve gained a matching defensive guard.
• linalg/traits/qr.rs: qr_solve_mut returns Err(Failed) ("Matrix is rank deficient.") for rank-deficient systems (duplicate/collinear columns, underdetermined shapes) instead of panicking.

[0.6.7]

Added

• linear/linear_regression.rs: native multi-output matrix support (#432, #433). LinearRegression::fit_matrix fits N x K targets directly (both QR and SVD solvers); predict_matrix returns the K-column prediction matrix; intercept_matrix exposes the 1 x K intercept row.

Changed

• Breaking: LinearRegression field intercept changed from Option<TX> to Option<X> (a 1 x K matrix) and the struct gained PhantomData<TX>. Previously serialized models will not deserialize; re-fit or migrate saved models. intercept() still returns a scalar for single-output callers.

[0.6.4]

Added

• Stage 2 test-coverage push (#393): proptest dev-dependency + property-based invariant tests and linalg edge cases.
  • linalg/basic/arrays.rs: proptest invariants — transpose involution (A^T)^T == A, matmul with identity A*I == A, matmul associativity (AB)C ≈ A(BC) (approximate comparison for FP), (AB)^T == B^T A^T, reshape preserves element count. Edge cases — 1x1 matmul, row×col matmul, shape-mismatch panic, reshape-incompatible panic, 1xN transpose.
  • algorithm/sort/quick_sort.rs: proptest — quick_argsort produces a valid permutation (all indices present exactly once, values non-decreasing in permutation order).
  • metrics/distance/euclidian.rs: proptest — d(a,a) == 0, symmetry d(a,b) == d(b,a), triangle inequality d(a,c) ≤ d(a,b) + d(b,c).

Changed

• Added proptest = "1.5" to [dev-dependencies].

[0.6.3]

Changed

• Replaced the remaining 4 unsafe {} raw-pointer blocks in linalg/basic/matrix.rs::iterator_mut / DenseMatrixMutView::iter_mut with a safe split_first_mut-based helper ordered_iter_mut (#368). The traversal order and offset formula are identical to the previous raw-pointer implementation; only the borrow-proving mechanism changed — eliminating unsafe from library code entirely.
• Performance note: the cross-axis path (axis ≠ natural storage order) now extracts the needed refs via split_first_mut in sorted-offset order and reorders, introducing a small allocation. The fast path (axis matches storage order) shortcuts to values.iter_mut().take(n) with zero overhead. Benchmarks to quantify the cross-axis delta are tracked in #407.

[0.6.2]

Changed

• Ported the crate from Rust edition 2021 to edition 2024 (#401, #402). cargo fix --edition made no auto-edits; the only behavioral-adjacent change is linalg/basic/arrays.rs::approximate_eq, rewritten to the 2024-safe tail-expr drop-order form (bind the owned intermediate before the borrowing iterator) — numerical logic unchanged.
• Declared rust-version = "1.85" (MSRV) in Cargo.toml and added an msrv CI job that builds with dtolnay/[email protected] to verify the claim (#404).
• Migrated lint suppressions: #[allow(...)] → #[expect(...)] at sites where the lint still fires under --all-features; #[allow] retained where the lint genuinely does not fire (avoids unfulfilled_lint_expectations) (#403).
• Added [lints.rust] unexpected_cfgs check-cfg table in Cargo.toml for cfg(coverage, coverage_nightly) and cfg(tarpaulin) (edition-2024 unexpected_cfgs lint).
• AGENTS.md: documented the edition-2024 invariants (no RPIT, explicit dyn Trait + 'a, tail-expr drop-order, lint-suppression policy, unsafe stance) and the "preserve bespoke numerical-system logic and performance" constraint for non-behavioral refactors.

Fixed

• svm/svc.rs: removed two redundant let svc = ...; svc tail expressions surfaced by the edition-2024 clippy::let_and_return lint.
• preprocessing/categorical.rs: kept the nested-if form (annotated #[allow(clippy::collapsible_if)]) because collapsing to a let-chain requires let-chains, unstable until Rust 1.88 — incompatible with the declared MSRV 1.85.

[0.6.1]

Added

• Stage 1 test-coverage push (#392): tests for previously-untested modules.
  • linalg/traits/high_order.rs: implemented the /* TODO */ test module — all 4 ab transpose-flag branches, non-square inputs, and a matmul/transpose equivalence check.
  • linear/lasso_optimizer.rs: direct tests for InteriorPointOptimizer (new shape of ata, known-answer l1-regularized least squares with lambda → 0).
  • error/mod.rs: tests for all 6 Failed constructors, all 8 FailedError variants, both Display impls, both PartialEq impls, and the Error trait impl.
  • rand_custom.rs: seeded-RNG determinism and None-seed usability tests.

Changed

• Revived 6 previously-commented-out serde round-trip tests (migrated to postcard, the post-#390 serialization backend) for LinearRegression, RidgeRegression, Lasso, ElasticNet, PCA, SVD.
• Fixed a latent type mismatch in the revived SVD serde test: the original commented-out code deserialized into SVD<f32, DenseMatrix<f32>> but SVD::fit on the f64 iris literals produces SVD<f64, ...> — corrected to SVD<f64, DenseMatrix<f64>>.
• Renamed two copy-paste-misnamed tests: dataset::diabetes::boston_dataset → diabetes_dataset; algorithm::sort::quick_sort::with_capacity → quick_argsort.

[0.6.0]

Changed

• CI coverage workflow now includes doctests and enforces a strict 44% line-coverage gate via cargo-tarpaulin (#399).

[0.5.3]

Changed

• Classification metrics refactored: Precision, Recall, and F1 now derive per-class scores from a single shared ConfusionCounts helper (src/metrics/confusion.rs) instead of each re-implementing the per-class tp/predicted/support bookkeeping. Precision and Recall expose a crate-private per_class_scores_from_counts used by F1's multiclass path.
• Precision and Recall now early-return 0.0 on empty input and drop the unreachable classes == 0 / support.is_empty() branches.
• Multiclass macro F1 (landed in #382, cleaned up in #383) is unchanged behaviourally; it now consumes Precision/Recall::per_class_scores_from_counts instead of its own HashMap bookkeeping.

[0.4.8] - 2025-11-29

• WARNING: Breaking changes!
• LassoParameters and LassoSearchParameters have a new field fit_intercept. When it is set to false, the beta_0 term in the formula will be forced to zero, and intercept field in Lasso will be set to None.

[0.4.0] - 2023-04-05

Added

• WARNING: Breaking changes!
• DenseMatrix constructor now returns Result to avoid user instantiating inconsistent rows/cols count. Their return values need to be unwrapped with unwrap(), see tests

[0.3.0] - 2022-11-09

Added

• WARNING: Breaking changes!
• Complete refactoring with extensive API changes that includes:
  • moving to a new traits system, less structs more traits
  • adapting all the modules to the new traits system
  • moving to Rust 2021, use of object-safe traits and as_ref
  • reorganization of the code base, eliminate duplicates
• implements readers (needs "serde" feature) for read/write CSV file, extendible to other formats
• default feature is now Wasm-/Wasi-first

Changed

• WARNING: Breaking changes!
• Seeds to multiple algorithims that depend on random number generation
• Added a new parameter to train_test_split to define the seed
• changed use of "serde" feature

Dropped

• WARNING: Breaking changes!
• Drop nalgebra-bindings feature, only ndarray as supported library

[0.2.1] - 2021-05-10

Added

• L2 regularization penalty to the Logistic Regression
• Getters for the naive bayes structs
• One hot encoder
• Make moons data generator
• Support for WASM.

Changed

• Make serde optional

[0.2.0] - 2021-01-03

Added

• DBSCAN
• Epsilon-SVR, SVC
• Ridge, Lasso, ElasticNet
• Bernoulli, Gaussian, Categorical and Multinomial Naive Bayes
• K-fold Cross Validation
• Singular value decomposition
• New api module
• Integration with Clippy
• Cholesky decomposition

Changed

• ndarray upgraded to 0.14
• smartcore::error:FailedError is now non-exhaustive
• K-Means
• PCA
• Random Forest
• Linear and Logistic Regression
• KNN
• Decision Tree

[0.1.0] - 2020-09-25

Added

• First release of smartcore.
• KNN + distance metrics (Euclidian, Minkowski, Manhattan, Hamming, Mahalanobis)
• Linear Regression (OLS)
• Logistic Regression
• Random Forest Classifier
• Decision Tree Classifier
• PCA
• K-Means
• Integrated with ndarray
• Abstract linear algebra methods
• RandomForest Regressor
• Decision Tree Regressor
• Serde integration
• Integrated with nalgebra
• LU, QR, SVD, EVD
• Evaluation Metrics