A simple true/false sampling tool for sampling n_times, n trials per epoch, with a given probability of success (defaults to 50%).

example.md

Try something twice, with a probability of failure at 10%, record wether the cummulative result was failure. Do that 1 million times, and form a posterior belief about this situation.

TruthSampling.call(n_samples: 1_000_000, trials: 2, probability: 0.1)

Or, here's a basketball example. If I've got a 1/6 free throw average, and I take 4 free throws, what is the chance I make at least one of those 4?

TruthSampling.call(trials: 4, probability: 1/6.0)
{false=>0.48256, true=>0.51744}

This gives me about a 52% chance of making at least one basket.

sampling.rb

class TruthSampling

  def self.call(opts={})
    new(opts).call
  end

  attr_reader :opts

  def initialize(opts={})
    @opts = opts
  end

  def n_samples
    @n_samples ||= opts.fetch(:n_samples, 100_000)
  end

  def trials
    @trials ||= opts.fetch(:trials, 10)
  end

  def probability
    @probability ||= opts.fetch(:probability, 0.5).to_f
  end

  def normalize(hash)
    total = hash.values.reduce(&:+).to_f
    hash.reduce({}) do |h, (k, v)|
      h[k] = v / total
      h
    end
  end

  def distribution
    n_samples.times.reduce(Hash.new(0)) do |hash, _|
      hash[sample] += 1
      hash
    end
  end

  def sample
    trials.times.any? { probe }
  end

  def probe
    rand <= probability
  end

  def call
    normalize(distribution)
  end

  def inspect
    "#{self.class.name} n_samples: #{n_samples}, trials: #{trials}, probability: #{probability}"
  end
end