SMath 1.4.0

<p align="center"> <img src="src/code/SMath/icon.png" alt="SMath" width="50"/> </p>

SMath

Geometry and statistics for .NET, written against generic math. Every formula is generic over the numeric type, so the same call site works for double, float, decimal or Half without overloads or casting.

Design

Generic over the number type. Constraints express the mathematics rather than a concrete type: a formula needing a square root asks for IRootFunctions<N>, one needing only addition asks for INumberBase<N>. Passing a type that cannot support the operation is a compile error.

Static and allocation free. There are no wrapper structs for points or vectors. Coordinates are plain tuples such as (N X, N Y), so values stay on the stack and interoperate with any other library. All entry points are static.

Names that read as the formula. Types are nested by the quantity being computed and the input it comes from, giving call sites like Circle.Region.Area.FromRadius(r) or Line.Segment.Length.FromTwoPoints(a, b).

Span overloads on the hot paths. Statistics and grid traversal accept ReadOnlySpan<T> next to IEnumerable<T>, avoiding enumerator allocation and interface dispatch per element. Coordinate generators additionally offer buffer filling overloads, so a caller can size a buffer with the matching count helper and reuse or stackalloc it.

Contents

Distance metrics available on Point2 and Point3: Euclidean, Manhattan, Chebyshev and Minkowski, plus Canberra in 3D.

Both points also generate integer coordinates by metric: neighbors, the coordinates at an exact distance, up to a distance or within a distance range, optionally limited by bounds. In 2D these are the taxicab circle and disk and the Chebyshev ring and square, in 3D the taxicab sphere and ball, an octahedron, and the Chebyshev shell and cube.

Setup

<PackageReference Include="SMath" Version="X.X.X" />

Replace X.X.X with the current version from NuGet. The package targets net7.0 and runs on any newer runtime.

Usage

Geometry, with the numeric type inferred from the arguments:

using SMath.Geometry2D;

// double and float from the same generic method
var tangentD = Circle.TangentLine.FromAngle(radius: 5d, angle: double.Pi / 4d);
var tangentF = Circle.TangentLine.FromAngle(radius: 5f, angle: float.Pi / 4f);

// tangent points from an external point, then the secant line through them
var points = Circle.TangentPoint.FromPoint(radius: 2d, (4, 4));
var secant = Line.FromTwoPoints(points.Value.Point1, points.Value.Point2);

var area = Circle.Region.Area.FromRadius(2d);

Statistics, over a sequence or a span:

using SMath.Statistics;

double[] values = [1, 2, 3, 4, 5];

var mean = ArithmeticMean.Eval(values);
var variance = Variance.Sample.Eval(values);
var deviation = StandardDeviation.Sample.Eval(values);

double[] other = [2, 1, 4, 3, 5];
var correlation = PearsonCorrelation.Eval<double>(values, other);

Grid traversal without allocating, using a count helper to size the buffer:

using SMath.Geometry2D;

var center = (X: 0, Y: 0);
const int radius = 3;

Span<(int X, int Y)> buffer = stackalloc (int X, int Y)[Point2.ManhattanDiskCount(radius)];
var count = Point2.CoordinatesUpToManhattanDistance(center, radius, buffer);

foreach (var coordinate in buffer[..count])
{
    // visit each coordinate within the taxicab disk
}

The same in three dimensions, walking a solid octahedron of voxels:

using SMath.Geometry3D;

var origin = (X: 0, Y: 0, Z: 0);

Span<(int X, int Y, int Z)> voxels = stackalloc (int X, int Y, int Z)[Point3.ManhattanBallCount(radius)];
var voxelCount = Point3.CoordinatesUpToManhattanDistance(origin, radius, voxels);

// only the shell, or the 26 neighbors of a voxel
var shell = Point3.CoordinatesAtChebyshevDistance(origin, radius);
var neighbors = Point3.AxialNeighbors(origin)
    .Concat(Point3.EdgeNeighbors(origin))
    .Concat(Point3.DiagonalNeighbors(origin));

Cross correlation over a range of lags, writing into a caller owned buffer:

using SMath.Statistics;

int[] lags = [-2, -1, 0, 1, 2];
var coefficients = new double[lags.Length];

PearsonCorrelation.Cross.Eval<double, int>(values, other, lags, coefficients);

Contributing

Ideas, bug reports and pull requests are welcome. Open an issue to propose a change, or send a pull request directly.

License

Project is under MIT license.