lernen 0.2.0 → 0.3.0

data/lib/lernen/algorithm/lsharp/lsharp_learner.rb

@@ -0,0 +1,367 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+module Lernen
+  module Algorithm
+    module LSharp
+      # LSharpLearner is an implementation of L# algorithm.
+      #
+      # L# is introduced by [Vaandrager et al. (2022) "A New Approach for Active
+      # Automata Learning Based on Apartness"](https://link.springer.com/chapter/10.1007/978-3-030-99524-9_12).
+      #
+      # @rbs generic In  -- Type for input alphabet
+      # @rbs generic Out -- Type for output values
+      class LSharpLearner < Learner #[In, Out]
+        # @rbs @alphabet: Array[In]
+        # @rbs @sul: System::SUL[In, Out]
+        # @rbs @oracle: Equiv::Oracle[In, Out]
+        # @rbs @automaton_type: :dfa | :mealy | :moore
+        # @rbs @tree: ObservationTree[In, Out]
+        # @rbs @witness_cache: Hash[[Array[In], Array[In]], Array[In]]
+        # @rbs @basis: Array[Array[In]]
+        # @rbs @frontier: Hash[Array[In], Array[Array[In]]]
+        # @rbs @incomplete_basis: Array[Array[In]]
+
+        #: (
+        #    Array[In] alphabet,
+        #    System::SUL[In, Out] sul,
+        #    automaton_type: :dfa | :mealy | :moore,
+        #  ) -> void
+        def initialize(alphabet, sul, automaton_type:)
+          super()
+
+          @alphabet = alphabet.dup
+          @sul = sul
+          @automaton_type = automaton_type
+
+          @tree = ObservationTree.new(sul, automaton_type:)
+          @witness_cache = {}
+
+          @basis = []
+          @basis_set = Set.new
+          @frontier = {}
+
+          @incomplete_basis = []
+
+          add_basis([])
+        end
+
+        # @rbs override
+        def build_hypothesis
+          loop do
+            next if promotion || completion || identification
+
+            hypothesis, state_to_prefix = build_hypothesis_internal
+            cex = check_consistency(hypothesis, state_to_prefix)
+            if cex
+              process_cex(cex, hypothesis, state_to_prefix)
+              update_frontier
+              next
+            end
+
+            return hypothesis, state_to_prefix
+          end
+
+          raise "BUG: unreachable"
+        end
+
+        # @rbs override
+        def refine_hypothesis(cex, hypothesis, state_to_prefix)
+          @tree.query(cex)
+
+          node = @tree.root
+          state = hypothesis.initial_conf
+          cex.size.times do |n|
+            input = cex[n]
+            node = node.branch[input]
+
+            _, state = hypothesis.step(state, input)
+            state_node = @tree[state_to_prefix[state]]
+            raise "BUG: A node for the basis prefix must exist" unless state_node
+
+            if check_apartness(state_node, node)
+              cex = cex[0..n]
+              break
+            end
+          end
+
+          process_cex(cex, hypothesis, state_to_prefix)
+        end
+
+        # @rbs override
+        def add_alphabet(input)
+          @alphabet << input
+          @incomplete_basis = @basis.dup
+        end
+
+        private
+
+        # Checks apartness on the current observation tree between the given two nodes.
+        # It returns the witness suffix if they are apart. If it is not, it returns `nil`.
+        #
+        #: (
+        #    ObservationTree::Node[In, Out] node1,
+        #    ObservationTree::Node[In, Out] node2
+        #  ) -> (Array[In] | nil)
+        def check_apartness(node1, node2)
+          case @automaton_type
+          in :dfa | :moore
+            return [] if node1.output != node2.output
+          in :mealy
+            # nop
+          end
+
+          queue = []
+          queue << [[], node1, node2]
+
+          until queue.empty?
+            suffix, node1, node2 = queue.shift
+            node1.branch.each do |input, next_node1|
+              next_node2 = node2.branch[input]
+              next unless next_node2
+              return suffix + [input] if next_node1.output != next_node2.output # steep:ignore
+              queue << [suffix + [input], next_node1, next_node2]
+            end
+          end
+
+          nil
+        end
+
+        # Computes the witness suffix of apartness between the given two basis prefixes.
+        #
+        #: (
+        #    Array[In] prefix1,
+        #    Array[In] prefix2
+        #  ) -> Array[In]
+        def compute_witness(prefix1, prefix2)
+          cached = @witness_cache[[prefix1, prefix2]]
+          return cached if cached
+
+          node1 = @tree[prefix1]
+          node2 = @tree[prefix2]
+          raise "BUG: Nodes for the basis prefixes must exist" unless node1 && node2
+
+          witness = check_apartness(node1, node2)
+          raise "BUG: A witness prefix for two basis prefixes must exist" unless witness
+
+          @witness_cache[[prefix1, prefix2]] = witness
+
+          witness
+        end
+
+        #: (Array[In] prefix) -> void
+        def add_basis(prefix)
+          @basis << prefix
+          @incomplete_basis << prefix
+          prefix_node = @tree[prefix]
+          raise "BUG: A node for the basis prefix must exist" unless prefix_node
+
+          @frontier.each do |border, eq_prefixes|
+            border_node = @tree[border]
+            raise "BUG: A node for the frontier prefix must exist" unless border_node
+
+            eq_prefixes << prefix unless check_apartness(prefix_node, border_node)
+          end
+        end
+
+        #: (Array[In] border) -> void
+        def add_frontier(border)
+          border_node = @tree[border]
+          raise "BUG: A node for the frontier prefix must exist" unless border_node
+
+          @frontier[border] = @basis.filter do |prefix|
+            prefix_node = @tree[prefix]
+            raise "BUG: A node for the basis prefix must exist" unless prefix_node
+
+            check_apartness(prefix_node, border_node).nil?
+          end
+        end
+
+        #: () -> void
+        def update_frontier
+          @frontier.each do |border, eq_prefixes|
+            border_node = @tree[border]
+            raise "BUG: A node for the frontier prefix must exist" unless border_node
+
+            eq_prefixes.filter! do |prefix|
+              prefix_node = @tree[prefix]
+              raise "BUG: A node for the basis prefix must exist" unless prefix_node
+
+              check_apartness(prefix_node, border_node).nil?
+            end
+          end
+        end
+
+        #: () -> [Automaton::TransitionSystem[Integer, In, Out], Hash[Integer, Array[In]]]
+        def build_hypothesis_internal
+          transitions = {}
+          prefix_to_state = @basis.each_with_index.to_h
+
+          @basis.each do |prefix|
+            state = prefix_to_state[prefix]
+            node = @tree[prefix]
+            raise "BUG: A node for the basis prefix must exist" unless node
+
+            @alphabet.each do |input|
+              next_node = node.branch[input]
+              next_prefix = prefix + [input]
+              next_state =
+                (
+                  if @frontier.include?(next_prefix)
+                    prefix_to_state[@frontier[next_prefix].first]
+                  else
+                    prefix_to_state[next_prefix]
+                  end
+                )
+
+              case @automaton_type
+              in :dfa | :moore
+                transitions[[state, input]] = next_state
+              in :mealy
+                transitions[[state, input]] = [next_node.output, next_state]
+              end
+            end
+          end
+
+          state_to_prefix = prefix_to_state.to_h { |q, i| [i, q] }
+          automaton =
+            case @automaton_type
+            in :dfa
+              accept_states =
+                state_to_prefix
+                  .to_a
+                  .filter { |(_, prefix)| @tree.observed_query(prefix).last }
+                  .to_set { |(state, _)| state }
+              Automaton::DFA.new(0, accept_states, transitions)
+            in :moore
+              outputs = state_to_prefix.transform_values { |state| @tree.observed_query(state).last }
+              Automaton::Moore.new(0, outputs, transitions)
+            in :mealy
+              Automaton::Mealy.new(0, transitions)
+            end
+
+          [automaton, state_to_prefix]
+        end
+
+        #: (
+        #    Automaton::TransitionSystem[Integer, In, Out] hypothesis,
+        #    Hash[Integer, Array[In]] state_to_prefix
+        #  ) -> (Array[In] | nil)
+        def check_consistency(hypothesis, state_to_prefix)
+          queue = []
+          queue << [[], hypothesis.initial_conf, @tree.root]
+
+          until queue.empty?
+            prefix, state, node = queue.shift
+            state_prefix = state_to_prefix[state]
+            next unless state_prefix
+
+            state_node = @tree[state_prefix]
+            raise "BUG: A node for the basis prefix must exist" unless state_node
+            return prefix if check_apartness(node, state_node)
+
+            node.branch.each do |input, next_node|
+              _, next_state = hypothesis.step(state, input)
+              queue << [prefix + [input], next_state, next_node]
+            end
+          end
+
+          nil
+        end
+
+        #: () -> bool
+        def promotion
+          @frontier.each do |new_prefix, eq_prefixes|
+            next unless eq_prefixes.empty?
+
+            @frontier.delete(new_prefix)
+            add_basis(new_prefix)
+
+            return true
+          end
+
+          false
+        end
+
+        #: () -> bool
+        def completion
+          updated = false
+
+          while (prefix = @incomplete_basis.pop)
+            prefix_node = @tree[prefix]
+            raise "BUG: A node for the basis prefix must exist" unless prefix_node
+
+            @alphabet.each do |input|
+              border = prefix + [input]
+              next if prefix_node.branch[input] && (@basis.include?(border) || @frontier.include?(border))
+
+              @tree.query(border)
+              add_frontier(border)
+
+              updated = true
+            end
+          end
+
+          updated
+        end
+
+        #: () -> bool
+        def identification
+          @frontier.each do |border, eq_prefixes|
+            next unless eq_prefixes.size >= 2
+
+            prefix1 = eq_prefixes[0]
+            prefix2 = eq_prefixes[1]
+            witness = compute_witness(prefix1, prefix2)
+            @tree.query(border + witness)
+            update_frontier
+
+            return true
+          end
+
+          false
+        end
+
+        #: (
+        #    Array[In] cex,
+        #    Automaton::TransitionSystem[Integer, In, Out] hypothesis,
+        #    Hash[Integer, Array[In]] state_to_prefix,
+        #  ) -> void
+        def process_cex(cex, hypothesis, state_to_prefix)
+          border = @frontier.keys.find { cex[0..._1.size] == _1 }
+          raise ArgumentError, "A border must exist" unless border
+
+          while border.size < cex.size
+            _, state = hypothesis.run(cex)
+            state_node = @tree[state_to_prefix[state]]
+            node = @tree[cex]
+            raise "BUG: Nodes for the basis prefix and `cex` must exist" unless state_node && node
+
+            witness = check_apartness(state_node, node)
+            raise ArgumentError, "A witness prefix for `cex` and its hypothesis state prefix must exist" unless witness
+
+            mid = border.size + ((cex.size - border.size) / 2)
+            cex1 = cex[0...mid]
+            cex2 = cex[mid...]
+
+            _, state1 = hypothesis.run(cex1) # steep:ignore
+            state1_prefix = state_to_prefix[state1]
+            @tree.query(state1_prefix + cex2 + witness) # steep:ignore
+
+            state1_node = @tree[state1_prefix]
+            node1 = @tree[cex1] # steep:ignore
+            raise "BUG: Nodes for the basis prefix and the prefix of `cex` must exist" unless state1_node && node1
+
+            if check_apartness(state1_node, node1)
+              cex = cex1
+            else
+              cex = state1_prefix + cex2 # steep:ignore
+              border = @frontier.keys.find { cex[0..._1.size] == _1 }
+              raise ArgumentError, "A border must exist" unless border
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/lernen/algorithm/lsharp/observation_tree.rb

@@ -0,0 +1,115 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+module Lernen
+  module Algorithm
+    module LSharp
+      # ObservationTree is an implementation of observation tree data structure.
+      #
+      # This data structure is used for L# algorithm.
+      #
+      # @rbs generic In  -- Type for input alphabet
+      # @rbs generic Out -- Type for output values
+      class ObservationTree
+        # Node is a node of an observation tree.
+        #
+        # `output` can take `nil` for the root node on learning Mealy machine.
+        #
+        # @rbs skip
+        Node =
+          Data.define(:output, :branch) do
+            def to_s
+              "#{output&.inspect}(#{branch.map { |c, n| "#{c.inspect} -> #{n}" }.join(", ")})" # steep:ignore
+            end
+          end
+
+        # @rbs!
+        #   class Node[In, Out] < Data
+        #     attr_reader output: Out | nil
+        #     attr_reader branch: Hash[In, Node[In, Out]]
+        #     def self.[]: [In, Out] (
+        #       Out output,
+        #       Hash[In, Node[In, Out]] branch
+        #     ) -> Node[In, Out]
+        #   end
+
+        # @rbs @sul: System::SUL[In, Out]
+        # @rbs @automaton_type: :dfa | :mealy | :moore
+        # @rbs @root: Node[In, Out]
+
+        #: (
+        #    System::SUL[In, Out] sul,
+        #    automaton_type: :dfa | :mealy | :moore
+        #  ) -> void
+        def initialize(sul, automaton_type:)
+          @sul = sul
+          @automaton_type = automaton_type
+
+          case automaton_type
+          in :dfa | :moore
+            raise "MooreLikeSUL is required to learn DFA or Moore" unless sul.is_a?(System::MooreLikeSUL)
+            @root = Node[sul.query_empty, {}]
+          in :mealy
+            @root = Node[nil, {}]
+          end
+        end
+
+        attr_reader :root #: Node[In, Out]
+
+        # Returns a node for the given word.
+        # When the word (or its subword) is not observed, it returns `nil` instead.
+        #
+        #: (Array[In] word) -> (Node[In, Out] | nil)
+        def [](word)
+          node = @root
+          word.each do |input|
+            return nil unless node.branch[input]
+            node = node.branch[input]
+          end
+          node
+        end
+
+        # Returns an output sequence for the given word if it is observed.
+        # If it is not, it returns `nil` instead.
+        #
+        #: (Array[In] word) -> (Array[Out] | nil)
+        def observed_query(word)
+          return [@root.output] if word.empty? # steep:ignore
+
+          node = @root
+          outputs = []
+          word.each do |input|
+            node = node.branch[input]
+            return nil unless node
+
+            outputs << node.output
+          end
+
+          outputs
+        end
+
+        # Returns an output sequence for the given word.
+        # When the word is observed, it runs actual query over `sul`.
+        #
+        #: (Array[In] word) -> Array[Out]
+        def query(word)
+          outputs = observed_query(word)
+          return outputs if outputs
+
+          node = @root
+          inprogress_word = []
+          outputs = []
+          word.each do |input|
+            inprogress_word << input
+            output = @sul.query_last(inprogress_word)
+            outputs << output
+            node.branch[input] ||= Node[output, {}] # steep:ignore
+            node = node.branch[input]
+          end
+
+          outputs
+        end
+      end
+    end
+  end
+end

data/lib/lernen/algorithm/lsharp.rb

@@ -0,0 +1,43 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+require "lernen/algorithm/lsharp/observation_tree"
+require "lernen/algorithm/lsharp/lsharp_learner"
+
+module Lernen
+  module Algorithm
+    # LSharp provides an implementation of L# algorithm.
+    #
+    # L# is introduced by [Vaandrager et al. (2022) "A New Approach for Active
+    # Automata Learning Based on Apartness"](https://link.springer.com/chapter/10.1007/978-3-030-99524-9_12).
+    module LSharp
+      # Runs the L# algorithm and returns an inferred automaton.
+      #
+      #: [In] (
+      #    Array[In] alphabet,
+      #    System::SUL[In, bool] sul,
+      #    Equiv::Oracle[In, bool] oracle,
+      #    automaton_type: :dfa,
+      #    ?max_learning_rounds: Integer | nil
+      #  ) -> Automaton::DFA[In]
+      #: [In, Out] (
+      #    Array[In] alphabet,
+      #    System::SUL[In, Out] sul,
+      #    Equiv::Oracle[In, Out] oracle,
+      #    automaton_type: :mealy,
+      #    ?max_learning_rounds: Integer | nil
+      #  ) -> Automaton::Mealy[In, Out]
+      #: [In, Out] (
+      #    Array[In] alphabet,
+      #    System::SUL[In, Out] sul,
+      #    Equiv::Oracle[In, Out] oracle,
+      #    automaton_type: :moore,
+      #    ?max_learning_rounds: Integer | nil
+      #  ) -> Automaton::Moore[In, Out]
+      def self.learn(alphabet, sul, oracle, automaton_type:, max_learning_rounds: nil) # steep:ignore
+        learner = LSharpLearner.new(alphabet, sul, automaton_type:)
+        learner.learn(oracle, max_learning_rounds:)
+      end
+    end
+  end
+end

data/lib/lernen/algorithm/lstar/lstar_learner.rb

@@ -0,0 +1,49 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+module Lernen
+  module Algorithm
+    module LStar
+      # LStarLearner is an implementation of Angluin's L* algorithm.
+      #
+      # Angluin's L* is introduced by [Angluin (1987) "Learning Regular Sets from
+      # Queries and Counterexamples"](https://dl.acm.org/doi/10.1016/0890-5401%2887%2990052-6).
+      #
+      # @rbs generic In  -- Type for input alphabet
+      # @rbs generic Out -- Type for output values
+      class LStarLearner < Learner #[In, Out]
+        # @rbs @alphabet: Array[In]
+        # @rbs @oracle: Equiv::Oracle[In, Out]
+        # @rbs @table: ObservationTable[In, Out]
+
+        #: (
+        #    Array[In] alphabet, System::SUL[In, Out] sul,
+        #    automaton_type: :dfa | :moore | :mealy,
+        #    ?cex_processing: cex_processing_method | nil
+        #  ) -> void
+        def initialize(alphabet, sul, automaton_type:, cex_processing: :binary)
+          super()
+
+          @alphabet = alphabet.dup
+
+          @table = ObservationTable.new(@alphabet, sul, automaton_type:, cex_processing:)
+        end
+
+        # @rbs override
+        def build_hypothesis
+          @table.build_hypothesis
+        end
+
+        # @rbs override
+        def refine_hypothesis(cex, hypothesis, state_to_prefix)
+          @table.refine_hypothesis(cex, hypothesis, state_to_prefix)
+        end
+
+        # @rbs override
+        def add_alphabet(input)
+          @alphabet << input
+        end
+      end
+    end
+  end
+end