WeightedRandom.Backend offers a contract to all who implement it:

1. The main WeightedRandom package presents a novel interface for generating a list of probabilities or weights
2. The custom Backend module decides what to do with those probabilities once they are generated.

Using Backends

To use a different backend, you can either pass it in as an opt (see function docs), or you can put it in your config file:

config :weighted_random,
  backend: Your.Backend.Module

Developing a new backend

To create your own backend, copy and change an existing one (like WeightedRandom.Backend.WalkerAlias or WeightedRandom.Backend.Linear).

Your module must implement the callbacks listed below: preprocess/2, take/2, and optionally options/0.

The simplest backend might look like this.

defmodule My.Backend do
  use WeightedRandom.Backend

  defstruct [:list]

  @impl true
  def preprocess(probabilities, _opts) do
    struct(__MODULE__, %{list: probabilities})
  end

  @impl true
  def take(%__MODULE__{list: li}, count) do
    for _ <- 1..count do
      Enum.random(li)
    end
  end 

end

This backend doesn't actually apply probability, it is just a glorified Enum.rand(list) for now. But you can use it in any of the main functions that take an opt of backend

WeightedRandom.rand(0..10, weights, [backend: My.Backend])
=> 3

Testing a backend

Testing random numbers is really hard. Even when you use seeds, it can be very brittle.

The WeightedRandom.Utils.Analysis module will be very useful for ensuring that your results follow the expected probabilities.

defmodule My.BackendTest do
  use ExUnit.Case
  alias WeightedRandom.Utils.Analysis
  alias My.Backend, as: Mod

  test "should return accurate sample" do
    probabilities = [0.25, 0.25, 0.5]
    sample = WeightedRandom.rand_p(probabilities, backend: Mod, take: 1000)
    tolerance = 0.05

    assert Analysis.match_probability?(probabilities, sample, tolerance)
  end
end