lernen 0.2.0 → 0.3.0

data/lib/lernen/algorithm/kearns_vazirani_vpa/discrimination_tree_vpa.rb

@@ -0,0 +1,246 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+module Lernen
+  module Algorithm
+    module KearnsVaziraniVPA
+      # DiscriminationTreeVPA is a extended version of discrimination tree for VPA.
+      #
+      # @rbs generic In     -- Type for input alphabet
+      # @rbs generic Call   -- Type for call alphabet
+      # @rbs generic Return -- Type for return alphabet
+      class DiscriminationTreeVPA
+        # @rbs skip
+        Node = Data.define(:access, :suffix, :branch)
+        # @rbs skip
+        Leaf = Data.define(:prefix)
+
+        # @rbs!
+        #   type tree[In, Call, Return] = Node[In, Call, Return]
+        #                               | Leaf[In, Call, Return]
+        #
+        #   class Node[In, Call, Return] < Data
+        #     attr_reader access: Array[In | Call | Return]
+        #     attr_reader suffix: Array[In | Call | Return]
+        #     attr_reader branch: Hash[bool, tree[In, Call, Return]]
+        #     def self.[]: [In, Call, Return] (
+        #       Array[In | Call | Return] access,
+        #       Array[In | Call | Return] suffix,
+        #       Hash[bool, tree[In, Call, Return]] branch
+        #     ) -> Node[In, Call, Return]
+        #   end
+        #
+        #   class Leaf[In, Call, Return] < Data
+        #     attr_reader prefix: Array[In | Call | Return]
+        #     def self.[]: [In, Call, Return] (
+        #       Array[In | Call | Return] prefix
+        #     ) -> Leaf[In, Call, Return]
+        #   end
+
+        # @rbs @alphabet: Array[In]
+        # @rbs @call_alphabet: Array[Call]
+        # @rbs @return_alphabet: Array[Return]
+        # @rbs @sul: System::MooreLikeSUL[In | Call | Return, bool]
+        # @rbs @cex_processing: cex_processing_method
+        # @rbs @path_hash: Hash[Array[In | Call | Return], Array[bool]]
+        # @rbs @root: Node[In, Call, Return]
+
+        #: (
+        #    Array[In] alphabet,
+        #    Array[Call] call_alphabet,
+        #    Array[Return] return_alphabet,
+        #    System::MooreLikeSUL[In | Call | Return, bool] sul,
+        #    cex: Array[In],
+        #    cex_processing: cex_processing_method
+        #  ) -> void
+        def initialize(alphabet, call_alphabet, return_alphabet, sul, cex:, cex_processing:)
+          @alphabet = alphabet
+          @call_alphabet = call_alphabet
+          @return_alphabet = return_alphabet
+          @sul = sul
+          @cex_processing = cex_processing
+
+          @path_hash = {}
+
+          @root = Node[[], [], {}]
+
+          empty_out = sul.query_empty
+          @root.branch[empty_out] = Leaf[[]]
+          @path_hash[[]] = [empty_out]
+
+          cex_out = sul.query_last(cex)
+          @root.branch[cex_out] = Leaf[cex] # steep:ignore
+          @path_hash[cex] = [cex_out] # steep:ignore
+        end
+
+        # Constructs a hypothesis automaton from this discrimination tree.
+        #
+        #: () -> [Automaton::VPA[In, Call, Return], Hash[Integer, Array[In | Call | Return]]]
+        def build_hypothesis
+          transition_function = {}
+          return_transition_function = {}
+
+          queue = []
+          prefix_to_state = {}
+          state_to_prefix = {}
+
+          queue << []
+          prefix_to_state[[]] = prefix_to_state.size
+          state_to_prefix[state_to_prefix.size] = []
+
+          until queue.empty?
+            prefix = queue.shift
+            state = prefix_to_state[prefix]
+            @alphabet.each do |input|
+              word = prefix + [input]
+              next_prefix = sift(word)
+
+              unless prefix_to_state.include?(next_prefix)
+                queue << next_prefix
+                prefix_to_state[next_prefix] = prefix_to_state.size
+                state_to_prefix[state_to_prefix.size] = next_prefix
+              end
+
+              next_state = prefix_to_state[next_prefix]
+              transition_function[[state, input]] = next_state
+
+              found_states = prefix_to_state.values
+
+              return_transition_function.each do |(return_state, return_input), return_transition_guard|
+                return_prefix = state_to_prefix[return_state]
+                @call_alphabet.each do |call_input|
+                  word = prefix + [call_input] + return_prefix + [return_input]
+                  next_prefix = sift(word)
+
+                  unless prefix_to_state.include?(next_prefix)
+                    queue << next_prefix
+                    prefix_to_state[next_prefix] = prefix_to_state.size
+                    state_to_prefix[state_to_prefix.size] = next_prefix
+                  end
+
+                  next_state = prefix_to_state[next_prefix]
+                  return_transition_guard[[state, call_input]] = next_state
+                end
+              end
+
+              @return_alphabet.each do |return_input|
+                return_transition_guard = return_transition_function[[state, return_input]] = {}
+                found_states.each do |call_state|
+                  call_prefix = state_to_prefix[call_state]
+                  @call_alphabet.each do |call_input|
+                    word = call_prefix + [call_input] + prefix + [return_input]
+                    next_prefix = sift(word)
+
+                    unless prefix_to_state.include?(next_prefix)
+                      queue << next_prefix
+                      prefix_to_state[next_prefix] = prefix_to_state.size
+                      state_to_prefix[state_to_prefix.size] = next_prefix
+                    end
+
+                    next_state = prefix_to_state[next_prefix]
+                    return_transition_guard[[call_state, call_input]] = next_state
+                  end
+                end
+              end
+            end
+          end
+
+          accept_state_set =
+            state_to_prefix.to_a.filter { |(_, prefix)| @path_hash[prefix][0] }.to_set { |(state, _)| state }
+          automaton = Automaton::VPA.new(0, accept_state_set, transition_function, return_transition_function)
+
+          [automaton, state_to_prefix]
+        end
+
+        # Update this classification tree by the given `cex`.
+        #
+        #: (
+        #    Array[In | Call | Return] cex,
+        #    Automaton::VPA[In, Call, Return] hypothesis,
+        #    Hash[Integer, Array[In | Call | Return]] state_to_prefix
+        #  ) -> void
+        def refine_hypothesis(cex, hypothesis, state_to_prefix)
+          conf_to_prefix = ->(conf) do
+            prefix = []
+
+            conf.stack.each do |state, call_input|
+              prefix.concat(state_to_prefix[state])
+              prefix << call_input
+            end
+            prefix.concat(state_to_prefix[conf.state])
+
+            prefix
+          end
+
+          acex = CexProcessor::PrefixTransformerAcex.new(cex, @sul, hypothesis, conf_to_prefix)
+          n = CexProcessor.process(acex, cex_processing: @cex_processing)
+          old_prefix = cex[0...n]
+          new_input = cex[n]
+          new_suffix = cex[n + 1...]
+
+          _, old_conf = hypothesis.run(old_prefix) # steep:ignore
+          _, replace_conf = hypothesis.step(old_conf, new_input)
+
+          new_access_conf = Automaton::VPA::Conf[hypothesis.initial_state, replace_conf.stack] # steep:ignore
+          new_access = conf_to_prefix.call(new_access_conf)
+
+          old_state_prefix = state_to_prefix[old_conf.state] # steep:ignore
+          if @alphabet.include?(new_input) # steep:ignore
+            new_prefix = old_state_prefix + [new_input]
+          else
+            call_state, call_input = old_conf.stack.last # steep:ignore
+            call_prefix = state_to_prefix[call_state]
+            new_prefix = call_prefix + [call_input] + old_state_prefix + [new_input]
+          end
+          new_out = @sul.query_last(new_access + new_prefix + new_suffix) # steep:ignore
+
+          replace_prefix = state_to_prefix[replace_conf.state] # steep:ignore
+          replace_out = @sul.query_last(new_access + replace_prefix + new_suffix) # steep:ignore
+
+          replace_node_path = @path_hash[replace_prefix]
+          replace_node_parent = @root
+          replace_node = @root.branch[replace_node_path.first] # steep:ignore
+          replace_node_path[1..].each do |out| # steep:ignore
+            replace_node_parent = replace_node
+            replace_node = replace_node.branch[out] # steep:ignore
+          end
+
+          new_node = Node[new_access, new_suffix, {}] # steep:ignore
+          replace_node_parent.branch[replace_node_path.last] = new_node # steep:ignore
+
+          new_node.branch[new_out] = Leaf[new_prefix] # steep:ignore
+          @path_hash[new_prefix] = replace_node_path + [new_out]
+
+          new_node.branch[replace_out] = Leaf[replace_prefix] # steep:ignore
+          @path_hash[replace_prefix] = replace_node_path + [replace_out]
+        end
+
+        private
+
+        # Returns a prefix discriminated by `word`.
+        #
+        #: (Array[In | Call | Return] word) -> Array[In | Call | Return]
+        def sift(word)
+          node = @root
+          path = []
+
+          until node.is_a?(Leaf)
+            full_word = node.access + word + node.suffix
+
+            out = @sul.query_last(full_word)
+            path << out
+
+            unless node.branch.include?(out) # steep:ignore
+              node.branch[out] = Leaf[word] # steep:ignore
+              @path_hash[word] = path
+            end
+
+            node = node.branch[out] # steep:ignore
+          end
+
+          node.prefix # steep:ignore
+        end
+      end
+    end
+  end
+end

data/lib/lernen/algorithm/kearns_vazirani_vpa/kearns_vazirani_vpa_learner.rb

@@ -0,0 +1,89 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+module Lernen
+  module Algorithm
+    module KearnsVaziraniVPA
+      # KearnzVaziraniVPALearner is an implementation of Kearnz-Vazirani algorithm for VPA.
+      #
+      # The idea behind this implementation is described by [Isberner (2015) "Foundations
+      # of Active Automata Learning: An Algorithmic Overview"](https://eldorado.tu-dortmund.de/handle/2003/34282).
+      #
+      # @rbs generic In     -- Type for input alphabet
+      # @rbs generic Call   -- Type for call alphabet
+      # @rbs generic Return -- Type for return alphabet
+      class KearnsVaziraniVPALearner < Learner #[In | Call | Return, bool]
+        # @rbs @alphabet: Array[In]
+        # @rbs @call_alphabet: Array[Call]
+        # @rbs @return_alphabet: Array[Return]
+        # @rbs @sul: System::MooreLikeSUL[In | Call | Return, bool]
+        # @rbs @oracle: Equiv::Oracle[In | Call | Return, bool]
+        # @rbs @cex_processing: cex_processing_method
+        # @rbs @tree: DiscriminationTreeVPA[In, Call, Return] | nil
+
+        #: (
+        #    Array[In] alphabet, Array[Call] call_alphabet, Array[Return] return_alphabet,
+        #    System::MooreLikeSUL[In | Call | Return, bool] sul,
+        #    ?cex_processing: cex_processing_method
+        #  ) -> void
+        def initialize(alphabet, call_alphabet, return_alphabet, sul, cex_processing: :binary)
+          super()
+
+          @alphabet = alphabet.dup
+          @call_alphabet = call_alphabet.dup
+          @return_alphabet = return_alphabet.dup
+          @sul = sul
+          @cex_processing = cex_processing
+
+          @tree = nil
+        end
+
+        # @rbs override
+        def build_hypothesis
+          tree = @tree
+          return tree.build_hypothesis if tree
+
+          [build_first_hypothesis, { 0 => [] }]
+        end
+
+        # @rbs override
+        def refine_hypothesis(cex, hypothesis, state_to_prefix)
+          tree = @tree
+          if tree
+            tree.refine_hypothesis(cex, hypothesis, state_to_prefix) # steep:ignore
+            return
+          end
+
+          @tree =
+            DiscriminationTreeVPA.new(
+              @alphabet,
+              @call_alphabet,
+              @return_alphabet,
+              @sul,
+              cex:,
+              cex_processing: @cex_processing
+            )
+        end
+
+        private
+
+        # Constructs the first hypothesis VPA.
+        #
+        #: () -> Automaton::VPA[In, Call, Return]
+        def build_first_hypothesis
+          transition_function = {}
+          @alphabet.each { |input| transition_function[[0, input]] = 0 }
+
+          return_transition_function = {}
+          @return_alphabet.each do |return_input|
+            return_transition_guard = return_transition_function[[0, return_input]] = {}
+            @call_alphabet.each { |call_input| return_transition_guard[[0, call_input]] = 0 }
+          end
+
+          accept_state_set = @sul.query_empty ? Set[0] : Set.new
+          Automaton::VPA.new(0, accept_state_set, transition_function, return_transition_function)
+        end
+      end
+    end
+  end
+end

data/lib/lernen/algorithm/kearns_vazirani_vpa.rb

@@ -0,0 +1,35 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+require "lernen/algorithm/kearns_vazirani_vpa/discrimination_tree_vpa"
+require "lernen/algorithm/kearns_vazirani_vpa/kearns_vazirani_vpa_learner"
+
+module Lernen
+  module Algorithm
+    # KearnzVaziraniVPA provides an implementation of Kearnz-Vazirani algorithm for VPA.
+    #
+    # The idea behind this implementation is described by [Isberner (2015) "Foundations
+    # of Active Automata Learning: An Algorithmic Overview"](https://eldorado.tu-dortmund.de/handle/2003/34282).
+    module KearnsVaziraniVPA
+      # Runs Kearns-Vazirani algorithm for VPA and returns an inferred VPA.
+      #
+      #: [In, Call, Return] (
+      #    Array[In] alphabet, Array[Call] call_alphabet, Array[Return] return_alphabet,
+      #    System::MooreLikeSUL[In | Call | Return, bool] sul, Equiv::Oracle[In | Call | Return, bool] oracle,
+      #    ?cex_processing: cex_processing_method, ?max_learning_rounds: Integer | nil
+      #  ) -> Automaton::VPA[In, Call, Return]
+      def self.learn( # steep:ignore
+        alphabet,
+        call_alphabet,
+        return_alphabet,
+        sul,
+        oracle,
+        cex_processing: :binary,
+        max_learning_rounds: nil
+      )
+        learner = KearnsVaziraniVPALearner.new(alphabet, call_alphabet, return_alphabet, sul, cex_processing:)
+        learner.learn(oracle, max_learning_rounds:)
+      end
+    end
+  end
+end

data/lib/lernen/algorithm/learner.rb

@@ -0,0 +1,82 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+module Lernen
+  module Algorithm
+    # Learner is an abstraction of implementations of learning algorithms.
+    #
+    # Note that this class is *abstract*. We should implement the following method:
+    #
+    # - `#oracle`
+    # - `#refine(cex, hypothesis, state_to_prefix)`
+    # - `#build_hypothesis`
+    #
+    # @rbs generic In  -- Type for input alphabet
+    # @rbs generic Out -- Type for output values
+    class Learner
+      # Runs the learning algorithm and returns an inferred automaton.
+      #
+      # `max_learning_rounds` is a parameter for specifying the maximum number of iterations for learning.
+      # When `max_learning_rounds: nil` is specified, it means the algorithm only stops if the equivalent
+      # hypothesis is found.
+      #
+      #: (
+      #    Equiv::Oracle[In, Out] oracle,
+      #    ?max_learning_rounds: Integer | nil
+      #  ) -> Automaton::TransitionSystem[untyped, In, Out]
+      def learn(oracle, max_learning_rounds: nil)
+        hypothesis, state_to_prefix = build_hypothesis
+        cex = oracle.find_cex(hypothesis)
+        return hypothesis if cex.nil?
+
+        learning_rounds = 0
+        loop do
+          break if max_learning_rounds && learning_rounds == max_learning_rounds
+          learning_rounds += 1
+
+          refine_hypothesis(cex, hypothesis, state_to_prefix)
+
+          hypothesis, state_to_prefix = build_hypothesis
+          cex = oracle.find_cex(hypothesis)
+          break if cex.nil?
+        end
+
+        hypothesis
+      end
+
+      # rubocop:disable Lint/UnusedMethodArgument
+
+      # Adds the given `input` to the alphabet.
+      #
+      # In the default, this method raises `TypeError` as the learner does not support
+      # adding an input character to the alphabet.
+      #
+      #: (In input) -> void
+      def add_alphabet(input)
+        raise TypeError, "This learner does not support adding an input character to the alphabet"
+      end
+
+      # Refine the learning hypothesis by the given counterexample.
+      #
+      # This is an abstract method.
+      #
+      #: (
+      #    Array[In] cex,
+      #    Automaton::TransitionSystem[untyped, In, Out] hypothesis,
+      #    Hash[Integer, Array[In]] state_to_prefix
+      #  ) -> void
+      def refine_hypothesis(cex, hypothesis, state_to_prefix)
+        raise TypeError, "abstract method: `refine_hypothesis`"
+      end
+
+      # This is an abstract method.
+      #r
+      #: () -> [Automaton::TransitionSystem[untyped, In, Out], Hash[Integer, Array[In]]]
+      def build_hypothesis
+        raise TypeError, "abstract method: `build_hypothesis`"
+      end
+
+      # rubocop:enable Lint/UnusedMethodArgument
+    end
+  end
+end