foresight

Predict the next output of a process which can only output binary states.

If a process outputs only two states (random or predictable), it's sometimes useful to predict the next state the process would output.

Why?

Here is a usage application:

function isAwesome(input) {
    
    if (expensiveComputation(input))
        return true;

    if (expensiveOperation(input))
        return true;

    return false;
}

// can potentially reduce potential computations by predicting which algorithm would hit first
function betterIsAwesome(input) {
    
    var order = foresight.guess();
    var algorithms = [expensiveComputation, expensiveOperation];
    var n = 2;

    if (order === 'algo1') {
        while(n-- > 0) {
            if(algorithms[1-n](input)) {
                (n == 1) && foresight.move('algo1') || foresight.move('algo2');
                return true;
            }
        }
    } else {
        while(n-- > 0) {
            if(algorithms[n](input)) {
                (n == 1) && foresight.move('algo2') || foresight.move('algo1');
                return true;
            }
        }
    }

    return false;
}

Usage

$ npm install --save foresight

var foresightMaker = require('foresight');

var foresight = foresightMaker();

foresight.move('head');
var guess = foresight.guess();

API

foresightMaker(options)

A factory function that returns an instance of foresight object.

options

Type: object

options.moves

By default foresight uses coin-flipping semantics to capture actual moves and to predict the next move.

You may remap these values to something else. Once they're remapped, you must use the new values for foresight.move(actual). And as well as expect them as output of foresight.guess().

Example:

options.move = {
    head: 'left',
    tail: 'right'
}

Default:

options.move = {
    head: 'head',
    tail: 'tail',
    pass: 'pass'
}

foresight.move(actual)

Input the actual state output of the tracking process.

actual

Value: head, tail, or values mapped to head and tail.

NOTE: actual may not equal to pass (in default move semantics) or move mapped to pass. Only foresight may return pass; which indicates that it is uncertain of its guess.

foresight.guess()

Returns a guess of the next state that a process would output.

If foresight is uncertain of the next move, it would output pass (or remapped value of pass).

Otherwise, if foresight is certain of the next move, it would output head or tail (or remapped values).

Credit

Code is ported to a usable npm module from: http://www.loper-os.org/bad-at-entropy/manmach.html

I did not come up with the underlying algorithm. I just rewrapped it into nicer API.

Weaknesses

As demonstrated by someone, the algorithm is deterministic, and one can beat it. You should expect that the process you're tracking won't likely output the precise sequences of moves that foresight would guess incorrectly most of the time.

The algorithm is light enough to be used in favour of a PRNG in some applications. Especially when heavyweight prediction tools isn't a likely solution.

License

MIT