Opt: Scalable Optimisation for Humans

opt is an optimisation library designed to make writing scalable optimisation routines easy. It provides a unified API for optimising within a single thread, multiple threads, or across machines, meaning the choice of how to distribute work is up to you.

Just go get github.com/cshenton/opt to install.

Basic Usage

Opt uses natural evolution strategies optimisers under the hood, which enables a simple API. First, Search() against an optimiser to get a test point, then evaluate its score how you see fit, then Show() that score with the test seed to the optimiser. Keep going until you converge.

package main

import (
	"fmt"

	"github.com/cshenton/opt"
	"github.com/cshenton/opt/bench"
)

func main() {
	n := 10000

	op := opt.DefaultOptions
	op.LearningRate = 0.5
	o := opt.NewSNES(10, op)

	for i := 0; i < n; i++ {
		point, seed := o.Search()
		// Minimise the 10-dimensional sphere function
		score := -bench.Sphere(point)
		o.Show(score, seed)
	}

	final, _ := o.Search()
	fmt.Println(final)
}

Want to use more workers on the same machine to speed up evaluations? Just do it, opt optimisers are thread safe, and will simply block calls to Search and Show when important computations are happening.

See examples for some suggestions on how to use opt.

Choice of Optimizers

Right now, opt deals with optimisers that work with real-valued inputs.

Available

• SNES (separable natural evolution strategies), great for high (>100) dimensional problems.

Coming Soon

• xNES (exponential natural evolution strategies), great for tricky low (<100) dimensional problems.
• Adaptive learning rates for XNES, SNES, great for unfamiliar problems.
• Hyperprojection SNES
• Block-free SNES

ToDo

• Simpler options handling
• xNES
• More expensive benchmark function to demonstrate multithread perf.
• worker queue example
• rpc example
• benchmark runner.
• some introspection / stopping functions for SNES
• Travis, coveralls

Dependencies

• golang.org/x/exp/rand Rob Pike's experimental rand library.
  • <1ns to create a random source vs. 8.5µs in std lib.
  • This probably matters more for smaller optimisation problems.