Machine Learning

Implemented — as hand-rolled numeric examples in ordinary Karkain arithmetic under examples/10-machine-learning/, byte-identical on both engines.

The honest label matters here: Karkain has no tensor types, no autodiff, no ML framework callable from ``.kark``. The matrix/vector machinery built in Phases 71–78 exists as compiler/backend infrastructure, not as a language surface. These examples instead implement two classic learning algorithms with plain floats, arrays and loops — which is exactly the kind of program a Karkain 1.2.0 user should be able to write today.

examples/10-machine-learning/

File

Demonstrates

01_linear_regression.kark

Closed-form ordinary least squares over a synthetic point set

02_gradient_descent.kark

Iterative loss minimization over a convex 1-D surface

Linear regression (excerpt)

func main() {
    let xs = [1.0, 2.0, 3.0, 4.0]
    let ys = [2.0, 4.0, 6.0, 8.0]
    ...
}

Legend: 0.9 / 1.3 / 6.7 pins the learned slope, intercept and a prediction for the synthetic fit.

Run:

$ karkain run examples/10-machine-learning/01_linear_regression.kark

See also

AI — nearest-neighbour and linear-classifier geometry. Roadmap — where a real ML surface may land.