JOS.WeightedResult

Different algorithms implemented in C# to use for getting a weighted result back (think A/B testing).

Currently implemented algorithms

• Alias Method (Vose) Heavily inspired by this article by Keith Schwarz and this answer on Stack Overflow. As you can see in the tests, the max deviation is roughly 0.65% (depending on the sample size ofc, higher == better).

Usage

It's highly recommended that you only create one WeightedResultQuery<T> instance per type/data set, preferably you register it as a singleton in your favorite DI container or use it as a static field. We want to avoid creating AliasMethodVose instances frequently since it does some initialization in the constructor.

The below example will print different results based on the following probability:

• 1/10 - "template-1"
• 2/10 - "template-2"
• 7/10 - "template-3"

var data = new List<ProbabilityItem<string>>
{
    new ProbabilityItem<string>(1, "template-1", "Template 1"),
    new ProbabilityItem<string>(2, "template-2", "Template 2"),
    new ProbabilityItem<string>(7, "template-3", "Template 3")
};

var query = new WeightedResultQuery<string>(data);
for(var i = 0; i < 100; i++)
{
    Console.WriteLine(query.Execute());
}

The following was printed to the console.

Output

template-3 template-2 template-3 template-2 template-3 template-1 template-3 template-3 template-3 template-2 template-3 template-3 template-3 template-3 template-3 template-1 template-3 template-3 template-2 template-2 template-3 template-3 template-2 template-3 template-3 template-3 template-3 template-3 template-3 template-3 template-3 template-3 template-1 template-3 template-3 template-2 template-1 template-2 template-3 template-3 template-3 template-2 template-3 template-1 template-3 template-3 template-1 template-3 template-3 template-2 template-2 template-3 template-2 template-3 template-2 template-3 template-2 template-2 template-3 template-3 template-3 template-3 template-3 template-3 template-3 template-2 template-3 template-1 template-3 template-2 template-3 template-3 template-3 template-3 template-2 template-3 template-3 template-3 template-1 template-3 template-3 template-3 template-3 template-3 template-3 template-3 template-2 template-3 template-3 template-3 template-3 template-1 template-3 template-3 template-3 template-1 template-3 template-3 template-3 template-3

Summary

template-3: 71
template-2: 19
template-1: 10

The probability for each item will always be the sum of all items / probability. Example:

var data = new List<ProbabilityItem<string>>
{
    new ProbabilityItem<string>(1, "template-1", "Template 1"),
    new ProbabilityItem<string>(2, "template-2", "Template 2"),
    new ProbabilityItem<string>(7, "template-3", "Template 3")
};

// The sum of all probabilities is 10 (1 + 2 + 7).
// The probability for the first item is 1/10 (10%)
// The probability for the second item is 2/10 (20%)
// The probability for the third item is 7/10 (70%)

The WeightedResultQuery contains a Probabilities property that will show how the query is configured. Example:

var data = new List<ProbabilityItem<string>>
{
    new ProbabilityItem<string>(1, "template-1", "Template 1"),
    new ProbabilityItem<string>(2, "template-2", "Template 2"),
    new ProbabilityItem<string>(7, "template-3", "Template 3")
};
var query = new WeightedResultQuery<string>(data);
var probabilities = query.Probabilities;

foreach(var probability in probabilities)
{
    Console.WriteLine($"{probability.Percentage} - {probability.Description}");
}

Output:

10 - Template 1
20 - Template 2
70 - Template 3

Roadmap

• Add support for simple A/B testing in ASP.NET Core.
• Implement more algorithms, for example Roulette Wheel Selection.