Vorcyc.Mathematics 0.10.20

Vorcyc.Mathematics

High-performance .NET 10 math library: SIMD/parallel CPU numerics, signal processing, deep learning, machine learning, statistics, linear algebra, and calculus.

Version: 0.10.20 · TFM: net10.0 · NuGet: Vorcyc.Mathematics

中文

Vorcyc® Mathematics targets .NET 10+, emphasizing SIMD and parallel CPU performance, accurate numerics, and modular APIs for DSP, deep learning, and classical ML.

Minimum target: .NET 10.0

Features

• SIMD-accelerated CPU execution
• Parallel execution via ComputingContext (scalar / SIMD / parallel / auto)
• Generic math (INumber, spans, tensors)
• No GPU in the main package — CPU-only; ILGPU removed (see GPU policy)

Dependencies

• System.Numerics.Tensors 10.0.9

Modules

Core — arrays, constants, RNG, trigonometry, pinnable buffers.
→ wiki/wiki_en/Module_Core.md

ComputingContext — execution policy, ComputingScope; integrated into FFT, statistics, linear algebra, batch DL, MFCC, Mel frontends, and Trainer.Fit*.
→ wiki/wiki_en/Module_ComputingContext.md

Deep learning (0.9) — BatchTensor (NHWC), BatchSequential / Sequential, 20+ batch layers, Trainer.Fit*, ModelSerializer v3.
→ wiki/wiki_en/Module_DeepLearning.md

Linear algebra
→ wiki/wiki_en/Module_LinearAlgebra.md

Machine learning (0.9+) — classifiers, regressors, pipelines, trees/forests, boosting, CV, grid search; Isolation Forest + JSON snapshots (0.10.11) & GPR PredictStd (0.10.10); CurveFitting SIMD large-N StackOverflow fix (0.10.16); classic CurveFitter methods take CancellationToken (0.10.17); PCA<T> out-of-sample Transform (0.10.18).
→ wiki/wiki_en/Module_MachineLearning.md

Signal processing — Signal / SignalSegment as primary time-domain APIs; FFT (TransformToFrequencyDomain uses SoA FftButterflyFp32 from 0.10.12), generic Fft<T> / RealFft<T> / HilbertTransform<T> facades (0.10.18), Convolver<T> / OlaBlockConvolver<T> / ComplexConvolver<T> / Operation.Convolve<T> (0.10.19), empty zpk ZpToTf([]) → [1] (0.10.20), mode decomposition stays in T (0.10.18), IirFilter<T> / FirFilter<T> / ZiFilter<T> / FDA-in-T (0.10.18), Welch (AveragePeriodogram complete windows + ComputingContext from 0.10.13; SciPy onesided scale / periodic windows / spectrum helpers from 0.10.14), FrequencyDomain.IndexToFrequency uses FFT TransformLength (0.10.15), pink noise via Paul Kellet (PinkNoiseGenerator, 0.10.14), filters, MFCC, effects.
→ wiki/wiki_en/Module_SignalProcessing.md

Numerics — extended integer/float/complex/rational types.
→ wiki/wiki_en/Module_Numerics.md

Statistics — reductions, distributions, time series (Holt / Holt–Winters terminal state + ForecastHolt recurrence, 0.10.18), extreme values.
→ wiki/wiki_en/Module_Statistics.md

Calculus — integration, derivatives, ODEs, root finding, series.
→ wiki/wiki_en/Module_Calculus.md

Extension packages

FFTW (Vorcyc.Mathematics.Extensions.FFTW, 0.10.20) — native FFTW bindings for high-performance DFTs (package version aligned; no FFTW API change).
→ wiki/wiki_en/Module_Extensions_FFTW.md

Quick start

dotnet add package Vorcyc.Mathematics

dotnet build targets/dotNET10_Portable/dotNET10_Portable.csproj -c Release
dotnet run --project Examples/Core_example -c Release

dotnet run --project TESTS/core_module_test/core_module_test.csproj -c Release
dotnet run --project TESTS/DL_module_test/DL_module_test.csproj -c Release

Solution layout

Open Vorcyc.Mathematics.slnx in Visual Studio or build individual projects with dotnet.

Examples

Standalone projects under Examples/ (not shipped on NuGet):

→ wiki/wiki_en/Module_Examples.md

Documentation

The two wikis are maintained independently. This solution README is a separate entry point and does not replace the wikis.

License

See docs/License.txt.