News

Changelog

This changelog includes end-user changes only. For internal development notes, see the repository changelog.

[5.0.0]

Added

• Enabled the optional wgpu DirectX 12 backend for GPU-enabled Windows builds. DXC is loaded dynamically with an FXC fallback, avoiding eager imports of dxcompiler.dll and dxil.dll.
• Added GLES for Android GPU builds with a target-scoped wgpu feature; Windows continues to omit the GLES-only loader imports.
• Added grouped outputs arrays and nested cv options for batch, streaming, online, and prediction configuration.
• Added a grouped intervals option with residual-bootstrap intervals for Lowess, StreamingLowess, full-update OnlineLowess, and LowessResult.predict().

Changed

• Clarified that Batch residual_sd is 1.4826 * MAD, while Streaming reports the cumulative sample standard deviation of emitted residuals.
• Breaking change: replaced individual return_* output booleans with outputs: [...] for Batch, Streaming, Online, and prediction options.
• Breaking change: replaced flat interval options and predict()’s interval levels with intervals: { confidence, prediction, bootstrap }, and replaced flat CV options and cv.seed with cv: { method, k, fractions } plus one outer seed shared by CV and bootstrap. predict() has its own seed.

Fixed

• Fixed CPU Gaussian standard errors for fitted and queried values to use the full unbounded kernel support.
• Made even-sample medians and mean-absolute residual scaling overflow-resistant for large finite values, and centered Batch/Streaming R-squared accumulation to preserve one-ULP response variation at large offsets.
• Made Batch RMSE, MAE, and R-squared reductions scale-safe for large finite values, preserved bisquare downweighting when tuned scales exceed the numeric range, and normalized custom weights before summation in local and all-tied fits.
• Kept AIC finite when raw residual-square sums overflow, and made local/all-tied WLS invariant to common scaling of large finite case weights.
• Keep generated N-API package-version checks in index.js synchronized with package.json after builds and version bumps.
• Fixed global OLS fits treating predictor values with a large offset as degenerate; translated inputs now retain their fitted slope.
• Fixed fraction-1 global fits ignoring custom weights, including when Batch sorts observations by x.
• Fixed Batch missing = "drop" accepting custom weights with a length different from the original input; weights are validated before rows are dropped.
• Fixed standard errors for global weighted fits to account for observation weights and weighted prediction leverage.
• Fixed local Batch and retained-model prediction standard errors ignoring custom_weights; local SE moments now include the per-observation case weights.
• Fixed global fits with all-zero custom weights to honor the configured zero-weight fallback policy.
• Fixed fraction-1 global fits ignoring configured robustness iterations; they now reweight observations and report iterations used.
• Fixed Batch cross-validation candidate fits ignoring custom_weights; K-fold CV now rejects more folds than observations instead of returning zero scores.
• Fixed Streaming and Online accepting invalid auto_converge tolerances; Online also rejects invalid explicit delta values while retaining NaN as its default sentinel.
• Fixed GPU Batch fits misaligning fitted values with unsorted inputs; results now preserve input and requested sorted order.
• Fixed GPU Batch silently ignoring custom_weights; GPU fit and CV candidate kernels now apply them directly.
• Reduced GPU adapter buffer requirements from 30 storage/32 total buffer bindings to 7 storage/8 total per shader stage by using per-compute-pipeline resource layouts.
• Fixed Online incremental mode accepting positive delta and auto_converge settings it cannot use; unsupported combinations now error.
• Fixed grouped intervals discarding distinct confidence and prediction levels; each requested coverage is now applied independently.
• Fixed OnlineLowess and LowessResult.predict() ignoring the outputs option.
• Reject unknown output names, detect musl reliably for GPU installs, reject unsupported musl ARM, validate GPU addons, and preserve the fit_async() declaration during builds.
• Fixed the default boundary_policy ("extend") letting synthetic boundary points bias the shared robustness scale estimate used to reweight every point, compounding across robustness iterations.
• Improved agreement with R/Cleveland on sparse, asymmetric, and high-iteration fits by aligning robustness stopping, local-linear degeneracy handling, neighborhood traversal, delta interpolation, and weighted accumulation.
• Fixed zero-radius neighborhoods dropping tied observations or accumulating normalized weights in a different order under robust fits.
• Fixed Gaussian fits clipping the unbounded kernel to the neighbor window and flooring far-tail weights; all observations now contribute under the standard Gaussian formula.

4.1.0

Added

• Added retain_model and LowessResult.predict(newX, options) for prediction.
• Added return_derivative to SmoothOptions, StreamingOptions, and OnlineOptions.
• Added standard errors and confidence/prediction intervals to StreamingOptions and OnlineOptions; online intervals require update_mode: "full".

Fixed

• Changed the OnlineLowess default iterations from 3 to 0, matching non-robust incremental updates; robustness iterations require update_mode: "full".
• Fixed cv_seed silently accepting negative values by rejecting them before casting.

4.0.0

Added

• Added return_sorted to SmoothOptions and missing to SmoothOptions, StreamingSmoothOptions, and OnlineSmoothOptions.
• Added fastlowess.installGpu() to download a prebuilt GPU addon; the installer also accepts a path to a local GPU artifact.
• Expanded prebuilt GPU binaries to cover all nine platforms supported by the npm package.

Changed

• Breaking change: split SmoothOptions into SmoothOptions for batch fits, StreamingSmoothOptions, and OnlineSmoothOptions; batch-only options now produce a TypeScript error instead of being silently ignored by streaming and online fits.
• Breaking change: removed ineffective return_diagnostics, return_residuals, and parallel options from OnlineOptions.
• Breaking change: StreamingOptions.overlap now defaults dynamically to chunk_size / 10, clamped to [1, chunk_size - 10], rather than a fixed 500.
• Improved the Node.js API documentation, including the dynamic overlap default.

Fixed

• Fixed npm run build and npm run build:debug to produce the platform-specific native addon files expected by the package loader.
• Fixed installGpu() to survive native-addon rebuilds, download the filename expected by the loader, and replace existing files on Windows.

3.2.1

Added

• Added prebuilt targets for Linux ARM64 with musl and ARMv7 Linux with hard-float support.

Changed

• Clarified that results are returned in input order, even though the algorithm sorts internally.

3.2.0

Added

• Added a VERSION export so consumers can query the Node.js package version directly.

Changed

• Reorganized and expanded the Node.js documentation, including a dedicated GPU backend guide with hardware requirements and performance considerations.

Fixed

• Improved the docs homepage and fixed links in the generated API reference.
• Corrected the Handling Outliers example to use a fraction that downweights the injected outlier.

3.1.0

Changed

• Moved Node.js documentation to GitHub Pages with executable examples and updated the README with package-specific guidance.

Fixed

• Changed .build() configuration failures to return Status::InvalidArg instead of Status::GenericFailure.

3.0.0

Added

• Added an opt-in GPU backend through the gpu Cargo feature and fastlowess.installGpu(); using the installed GPU addon requires restarting Node.js.
• Added cross-reference links in the API documentation to the corresponding user guides.

Changed

• Breaking change: renamed OnlineOutput fields smoothed and std_error to y and standard_error.
• Organized Streaming and Online API documentation and tutorials into dedicated user-guide pages.

2.0.0

Added

• Added OnlineOutput for OnlineLowess.add_point(), exposing the smoothed value, standard error, residual, robustness weight, and iterations used.
• Added return_se and cv_seed options to SmoothOptions.
• Added customWeights to fit and fit_async for per-observation batch weights.
• Unknown option keys now throw a TypeError listing the valid keys.

Changed

• Breaking change: renamed public API fields, methods, and options from camelCase to snake_case.
• Breaking change: replaced OnlineLowess.add_points(x, y) with add_point(x, y), which processes one point and returns OnlineOutput | null.
• Changed OnlineOptions.window_capacity’s default from 100 to 1000 and min_points from 2 to 3.
• OnlineLowess now forwards all SmoothOptions fields to the underlying builder instead of silently ignoring most of them.

1.3.0

Fixed

• Changed configuration errors from .build() to return Status::InvalidArg rather than a generic runtime failure status.

1.2.0

Fixed

• Fixed reported vulnerabilities in Node.js dependencies.

1.1.2

• Under-the-hood maintenance; no changes to the public API or runtime behavior.

1.1.1

Fixed

• Fixed GPU configuration and initialization problems, and improved recovery after missing hardware/drivers or earlier GPU execution errors.

1.1.0

Added

• Expanded GPU fitting support to selectable kernels, robustness and scaling methods, boundary policies, automatic convergence, prediction, and cross-validation.

Fixed

• Fixed the Extend boundary policy not being applied and improved numerical precision through coordinate centering.
• Fixed GPU integer overflow, initialization failures, and resource exhaustion.

1.0.0

Added

• Added the mean scaling method (Mean Absolute Deviation).
• Added asynchronous batch processing.

0.99.9

Added

• Made the package available on npm as fastlowess.

Changed

• Introduced class-based builders for streaming and online processing.

0.99.8

• Initial implementation of the Node.js binding.