mikeg-vanity 1.3.0

data/lib/vanity/experiment/ab_test.rb

@@ -0,0 +1,477 @@
+require "digest/md5"
+
+module Vanity
+  module Experiment
+
+    # One of several alternatives in an A/B test (see AbTest#alternatives).
+    class Alternative
+
+      def initialize(experiment, id, value, participants, converted, conversions)
+        @experiment = experiment
+        @id = id
+        @name = "option #{(@id + 65).chr}"
+        @value = value
+        @participants, @converted, @conversions = participants, converted, conversions
+      end
+
+      # Alternative id, only unique for this experiment.
+      attr_reader :id
+     
+      # Alternative name (option A, option B, etc).
+      attr_reader :name
+
+      # Alternative value.
+      attr_reader :value
+
+      # Experiment this alternative belongs to.
+      attr_reader :experiment
+
+      # Number of participants who viewed this alternative.
+      attr_reader :participants
+
+      # Number of participants who converted on this alternative (a participant is counted only once).
+      attr_reader :converted
+
+      # Number of conversions for this alternative (same participant may be counted more than once).
+      attr_reader :conversions
+
+      # Z-score for this alternative, related to 2nd-best performing alternative. Populated by AbTest#score.
+      attr_accessor :z_score
+
+      # Probability derived from z-score. Populated by AbTest#score.
+      attr_accessor :probability
+    
+      # Difference from least performing alternative. Populated by AbTest#score.
+      attr_accessor :difference
+
+      # Conversion rate calculated as converted/participants, rounded to 3 places.
+      def conversion_rate
+        @conversion_rate ||= (participants > 0 ? (converted.to_f/participants.to_f * 1000).round / 1000.0 : 0.0)
+      end
+
+      # The measure we use to order (sort) alternatives and decide which one is better (by calculating z-score).
+      # Defaults to conversion rate.
+      def measure
+        conversion_rate
+      end
+
+      def <=>(other)
+        measure <=> other.measure 
+      end
+
+      def ==(other)
+        other && id == other.id && experiment == other.experiment
+      end
+
+      def to_s
+        name
+      end
+
+      def inspect
+        "#{name}: #{value} #{converted}/#{participants}"
+      end
+
+    end
+
+
+    # The meat.
+    class AbTest < Base
+      class << self
+
+        # Convert z-score to probability.
+        def probability(score)
+          score = score.abs
+          probability = AbTest::Z_TO_PROBABILITY.find { |z,p| score >= z }
+          probability ? probability.last : 0
+        end
+
+        def friendly_name
+          "A/B Test" 
+        end
+
+      end
+
+      def initialize(*args)
+        super
+        @alternatives = [false, true]
+      end
+
+
+      # -- Metric --
+    
+      # Tells A/B test which metric we're measuring, or returns metric in use. 
+      #
+      # @example Define A/B test against coolness metric
+      #   ab_test "Background color" do
+      #     metrics :coolness
+      #     alternatives "red", "blue", "orange"
+      #   end
+      # @example Find metric for A/B test
+      #   puts "Measures: " + experiment(:background_color).metrics.map(&:name)
+      def metrics(*args)
+        @metrics = args.map { |id| @playground.metric(id) } unless args.empty?
+        @metrics
+      end
+
+
+      # -- Alternatives --
+
+      # Call this method once to set alternative values for this experiment
+      # (requires at least two values).  Call without arguments to obtain
+      # current list of alternatives.
+      #
+      # @example Define A/B test with three alternatives
+      #   ab_test "Background color" do
+      #     metrics :coolness
+      #     alternatives "red", "blue", "orange"
+      #   end
+      # 
+      # @example Find out which alternatives this test uses
+      #   alts = experiment(:background_color).alternatives
+      #   puts "#{alts.count} alternatives, with the colors: #{alts.map(&:value).join(", ")}"
+      def alternatives(*args)
+        unless args.empty?
+          @alternatives = args.clone
+        end
+        class << self
+          define_method :alternatives, instance_method(:_alternatives)
+        end
+        alternatives
+      end
+
+      def _alternatives
+        alts = []
+        @alternatives.each_with_index do |value, i|
+          participants = redis.scard(key("alts:#{i}:participants")).to_i
+          converted = redis.scard(key("alts:#{i}:converted")).to_i
+          conversions = redis[key("alts:#{i}:conversions")].to_i
+          alts << Alternative.new(self, i, value, participants, converted, conversions)
+        end
+        alts
+      end
+      private :_alternatives
+
+      # Returns an Alternative with the specified value.
+      #
+      # @example
+      #   alternative(:red) == alternatives[0]
+      #   alternative(:blue) == alternatives[2]
+      def alternative(value)
+        alternatives.find { |alt| alt.value == value }
+      end
+
+      # Defines an A/B test with two alternatives: false and true.  This is the
+      # default pair of alternatives, so just syntactic sugar for those who love
+      # being explicit.
+      #
+      # @example
+      #   ab_test "More bacon" do
+      #     metrics :yummyness 
+      #     false_true
+      #   end
+      #
+      def false_true
+        alternatives false, true
+      end
+      alias true_false false_true
+
+      # Chooses a value for this experiment.  You probably want to use the
+      # Rails helper method ab_test instead.
+      #
+      # This method picks an alternative for the current identity and returns
+      # the alternative's value.  It will consistently choose the same
+      # alternative for the same identity, and randomly split alternatives
+      # between different identities.
+      #
+      # @example
+      #   color = experiment(:which_blue).choose
+      def choose
+        if active?
+          identity = identity()
+          index = redis[key("participant:#{identity}:show")]
+          unless index
+            index = alternative_for(identity)
+            redis.sadd key("alts:#{index}:participants"), identity
+            check_completion!
+          end
+        else
+          index = redis[key("outcome")] || alternative_for(identify)
+        end
+        @alternatives[index.to_i]
+      end
+
+      # Returns fingerprint (hash) for given alternative.  Can be used to lookup
+      # alternative for experiment without revealing what values are available
+      # (e.g. choosing alternative from HTTP query parameter).
+      def fingerprint(alternative)
+        Digest::MD5.hexdigest("#{id}:#{alternative.id}")[-10,10]
+      end
+
+      
+      # -- Testing --
+     
+      # Forces this experiment to use a particular alternative.  You'll want to
+      # use this from your test cases to test for the different alternatives.
+      #
+      # @example Setup test to red button
+      #   setup do
+      #     experiment(:button_color).select(:red)
+      #   end
+      #
+      #   def test_shows_red_button
+      #     . . .
+      #   end
+      #
+      # @example Use nil to clear selection
+      #   teardown do
+      #     experiment(:green_button).select(nil)
+      #   end
+      def chooses(value)
+        if value.nil?
+          redis.del key("participant:#{identity}:show")
+        else
+          index = @alternatives.index(value)
+          raise ArgumentError, "No alternative #{value.inspect} for #{name}" unless index
+          redis[key("participant:#{identity}:show")] = index
+        end
+        self
+      end
+
+      # True if this alternative is currently showing (see #chooses).
+      def showing?(alternative)
+        identity = identity()
+        index = redis[key("participant:#{identity}:show")]
+        index && index.to_i == alternative.id
+      end
+
+
+      # -- Reporting --
+
+      # Scores alternatives based on the current tracking data.  This method
+      # returns a structure with the following attributes:
+      # [:alts]   Ordered list of alternatives, populated with scoring info.
+      # [:base]   Second best performing alternative.
+      # [:least]  Least performing alternative (but more than zero conversion).
+      # [:choice] Choice alterntive, either the outcome or best alternative.
+      #
+      # Alternatives returned by this method are populated with the following
+      # attributes:
+      # [:z_score]      Z-score (relative to the base alternative).
+      # [:probability]  Probability (z-score mapped to 0, 90, 95, 99 or 99.9%).
+      # [:difference]   Difference from the least performant altenative.
+      # 
+      # The choice alternative is set only if its probability is higher or
+      # equal to the specified probability (default is 90%).
+      def score(probability = 90)
+        alts = alternatives
+        # sort by conversion rate to find second best and 2nd best
+        sorted = alts.sort_by(&:measure)
+        base = sorted[-2]
+        # calculate z-score
+        pc = base.measure
+        nc = base.participants
+        alts.each do |alt|
+          p = alt.measure
+          n = alt.participants
+          alt.z_score = (p - pc) / ((p * (1-p)/n) + (pc * (1-pc)/nc)).abs ** 0.5
+          alt.probability = AbTest.probability(alt.z_score)
+        end
+        # difference is measured from least performant
+        if least = sorted.find { |alt| alt.measure > 0 }
+          alts.each do |alt|
+            if alt.measure > least.measure
+              alt.difference = (alt.measure - least.measure) / least.measure * 100
+            end
+          end
+        end
+        # best alternative is one with highest conversion rate (best shot).
+        # choice alternative can only pick best if we have high probability (>90%).
+        best = sorted.last if sorted.last.measure > 0.0
+        choice = outcome ? alts[outcome.id] : (best && best.probability >= probability ? best : nil)
+        Struct.new(:alts, :best, :base, :least, :choice).new(alts, best, base, least, choice)
+      end
+
+      # Use the result of #score to derive a conclusion.  Returns an
+      # array of claims.
+      def conclusion(score = score)
+        claims = []
+        participants = score.alts.inject(0) { |t,alt| t + alt.participants }
+        claims << case participants
+          when 0 ; "There are no participants in this experiment yet."
+          when 1 ; "There is one participant in this experiment."
+          else ; "There are #{participants} participants in this experiment."
+        end
+        # only interested in sorted alternatives with conversion
+        sorted = score.alts.select { |alt| alt.measure > 0.0 }.sort_by(&:measure).reverse
+        if sorted.size > 1
+          # start with alternatives that have conversion, from best to worst,
+          # then alternatives with no conversion.
+          sorted |= score.alts
+          # we want a result that's clearly better than 2nd best.
+          best, second = sorted[0], sorted[1]
+          if best.measure > second.measure
+            diff = ((best.measure - second.measure) / second.measure * 100).round
+            better = " (%d%% better than %s)" % [diff, second.name] if diff > 0
+            claims << "The best choice is %s: it converted at %.1f%%%s." % [best.name, best.measure * 100, better]
+            if best.probability >= 90
+              claims << "With %d%% probability this result is statistically significant." % score.best.probability
+            else
+              claims << "This result is not statistically significant, suggest you continue this experiment."
+            end
+            sorted.delete best
+          end
+          sorted.each do |alt|
+            if alt.measure > 0.0
+              claims << "%s converted at %.1f%%." % [alt.name.gsub(/^o/, "O"), alt.measure * 100]
+            else
+              claims << "%s did not convert." % alt.name.gsub(/^o/, "O")
+            end
+          end
+        else
+          claims << "This experiment did not run long enough to find a clear winner."
+        end
+        claims << "#{score.choice.name.gsub(/^o/, "O")} selected as the best alternative." if score.choice
+        claims
+      end
+
+
+      # -- Completion --
+
+      # Defines how the experiment can choose the optimal outcome on completion.
+      #
+      # By default, Vanity will take the best alternative (highest conversion
+      # rate) and use that as the outcome.  You experiment may have different
+      # needs, maybe you want the least performing alternative, or factor cost
+      # in the equation?
+      #
+      # The default implementation reads like this:
+      #   outcome_is do
+      #     a, b = alternatives
+      #     # a is expensive, only choose a if it performs 2x better than b
+      #     a.measure > b.measure * 2 ? a : b
+      #   end
+      def outcome_is(&block)
+        raise ArgumentError, "Missing block" unless block
+        raise "outcome_is already called on this experiment" if @outcome_is
+        @outcome_is = block
+      end
+
+      # Alternative chosen when this experiment completed.
+      def outcome
+        outcome = redis[key("outcome")]
+        outcome && alternatives[outcome.to_i]
+      end
+
+      def complete!
+        return unless active?
+        super
+        if @outcome_is
+          begin
+            result = @outcome_is.call
+            outcome = result.id if result && result.experiment == self
+          rescue
+            # TODO: logging
+          end
+        else
+          best = score.best
+          outcome = best.id if best
+        end
+        # TODO: logging
+        redis.setnx key("outcome"), outcome || 0
+      end
+
+      
+      # -- Store/validate --
+
+      def destroy
+        @alternatives.size.times do |i|
+          redis.del key("alts:#{i}:participants")
+          redis.del key("alts:#{i}:converted")
+          redis.del key("alts:#{i}:conversions")
+        end
+        redis.del key(:outcome)
+        super
+      end
+
+      def save
+        fail "Experiment #{name} needs at least two alternatives" unless alternatives.size >= 2
+        super
+        if @metrics.nil? || @metrics.empty?
+          warn "Please use metrics method to explicitly state which metric you are measuring against."
+          metric = @playground.metrics[id] ||= Vanity::Metric.new(@playground, name)
+          @metrics = [metric]
+        end
+        @metrics.each do |metric|
+          metric.hook &method(:track!)
+        end
+      end
+
+      # Called when tracking associated metric.
+      def track!(metric_id, timestamp, count, *args)
+        return unless active?
+        identity = identity() rescue nil
+        if identity
+          return if redis[key("participants:#{identity}:show")]
+          index = alternative_for(identity)
+          redis.sadd key("alts:#{index}:converted"), identity if redis.sismember(key("alts:#{index}:participants"), identity)
+          redis.incrby key("alts:#{index}:conversions"), count
+          check_completion!
+        end
+      end
+
+      # If you are not embarrassed by the first version of your product, you’ve
+      # launched too late.
+      #   -- Reid Hoffman, founder of LinkedIn
+
+    protected
+
+      # Used for testing.
+      def fake(values)
+        values.each do |value, (participants, conversions)|
+          conversions ||= participants
+          participants.times do |identity|
+            index = @alternatives.index(value)
+            raise ArgumentError, "No alternative #{value.inspect} for #{name}" unless index
+            redis.sadd key("alts:#{index}:participants"), identity
+          end
+          conversions.times do |identity|
+            index = @alternatives.index(value)
+            raise ArgumentError, "No alternative #{value.inspect} for #{name}" unless index
+            redis.sadd key("alts:#{index}:converted"), identity
+            redis.incr key("alts:#{index}:conversions")
+          end
+        end
+      end
+
+      # Chooses an alternative for the identity and returns its index. This
+      # method always returns the same alternative for a given experiment and
+      # identity, and randomly distributed alternatives for each identity (in the
+      # same experiment).
+      def alternative_for(identity)
+        Digest::MD5.hexdigest("#{name}/#{identity}").to_i(17) % @alternatives.size
+      end
+
+      begin
+        a = 50.0
+        # Returns array of [z-score, percentage]
+        norm_dist = []
+        (0.0..3.1).step(0.01) { |x| norm_dist << [x, a += 1 / Math.sqrt(2 * Math::PI) * Math::E ** (-x ** 2 / 2)] }
+        # We're really only interested in 90%, 95%, 99% and 99.9%.
+        Z_TO_PROBABILITY = [90, 95, 99, 99.9].map { |pct| [norm_dist.find { |x,a| a >= pct }.first, pct] }.reverse
+      end
+
+    end
+
+
+    module Definition
+      # Define an A/B test with the given name.  For example:
+      #   ab_test "New Banner" do
+      #     alternatives :red, :green, :blue
+      #   end
+      def ab_test(name, &block)
+        define name, :ab_test, &block
+      end
+    end
+
+  end
+end