niML is eventually going to be a machine learning library for nimrod. Currently it only provides numerical optimization functionality.
Let's find the minimum of the function y(x) = (x-x_0)^2 for x_0 = (2,1) using gradient descent and Newton's method. For this we will have to pass the gradient and Hessian of the function.
import niml.opt, niml.vec
echo opt.gd(proc(x: array[2, float]) : auto = 2*(x-[2.0, 1.0]), [1.0, 0.0])
echo opt.nm(proc(x: array[2, float]) : auto = 2*(x-[2.0, 1.0]),
proc(x: array[2, float]) : auto = array([2.0, 2.0]), [1.0, 0.0])The result is
[ 1.9999976054757174e+00, 9.9999760547571737e-01]
[ 2.0000000000000000e+00, 1.0000000000000000e+00]
There's obviously a lot to do, and you can help.
- Comment the code.
- Write unit tests.
- Create a package.
- File bug reports.
- Write ML routines.
- Add multi-threading.
- Integrate PLASMA/MAGMA.
This software is released under the GNU GPL v2.0 license.