search-optimisation-streams

Last updated: 2016-03-19 17:08:22 +0000

Upstream URL: git clone http://chriswarbo.net/git/search-optimisation-streams.git

Repo

View repository

View issue tracker

Contents of follows


Search and Optimisation Streams

This library defines a bunch of streams (infinite lists), for search and optimisation.

A stream is a never-ending source of values. In our case, a stream is implemented as a Javascript function with no arguments. This is a "thunk", or a computation-in-waiting. Whenever we like, we can call the function and it returns the next value for us. Calling the function again will return the next value, and so on. The next value of a stream can depend arbitrarily on the previous values, but it cannot depend on any future values. This is enabled by making all streams either pure functions, or closures.

Search is the problem of constructing suitable inputs for a function, such that it gives a known output. This is somewhat the inverse of most computations, which take inputs and generate unknown outputs. Usually, as is the case in this library, the output we want is boolean true, and the function is some user-defined measure of acceptibility. We usually require the user supplies the domain of values as well (eg. booleans, integers, strings, etc.), or at least some enumerating function.

Optimisation is a generalisation of search. Instead of requiring a specific value, we use an ordered co-domain (for example, numbers), and construct values which give higher and higher acceptibility.

This library provides many basic streams (for example "zeros" and "ones", which give values 0, 0, 0, ... and 1, 1, 1, ... respectively), as well as a wealth of combinators to produce new streams (for example, interleave(zeros(), ones()) gives 0, 1, 0, 1, 0, 1, ...) and stream-building functions (for example constant, where constant(x) gives x, x, x, ...).

By using the common framework of streams, we build up from these basic building blocks to quite elaborate metaheuristic search/optimisation algorithms, including genetic algorithms and virtual machine enumerators.