Merge pull request #102 from Herb-AI/dev

v0.3

Project.toml

@@ -1,7 +1,7 @@
 name = "HerbSearch"
 uuid = "3008d8e8-f9aa-438a-92ed-26e9c7b4829f"
 authors = ["Sebastijan Dumancic <[email protected]>", "Jaap de Jong <[email protected]>", "Nicolae Filat <[email protected]>", "Piotr Cichoń <[email protected]>", "Tilman Hinnerichs <[email protected]>"]
-version = "0.2.0"
+version = "0.3.0"
 
 [deps]
 DataStructures = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"
@@ -17,10 +17,10 @@ StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
 
 [compat]
 DataStructures = "0.17,0.18"
-HerbConstraints = "^0.1.0"
-HerbCore = "^0.2.0"
-HerbGrammar = "^0.2.0"
-HerbInterpret = "^0.1.1"
+HerbConstraints = "^0.2.0"
+HerbCore = "^0.3.0"
+HerbGrammar = "^0.3.0"
+HerbInterpret = "^0.1.3"
 HerbSpecification = "^0.1.0"
 MLStyle = "^0.4.17"
 StatsBase = "^0.34"

src/HerbSearch.jl

@@ -12,10 +12,11 @@ using MLStyle
 include("sampling_grammar.jl")
 
 include("program_iterator.jl")
-include("count_expressions.jl")
+include("uniform_iterator.jl")
 
 include("heuristics.jl")
 
+include("fixed_shaped_iterator.jl")
 include("top_down_iterator.jl")
 
 include("evaluate.jl")
@@ -35,8 +36,9 @@ include("genetic_functions/crossover.jl")
 include("genetic_functions/select_parents.jl")
 include("genetic_search_iterator.jl")
 
+include("random_iterator.jl")
+
 export 
-  count_expressions,
   ProgramIterator,
   @programiterator,
 
@@ -47,12 +49,19 @@ export
   heuristic_random,
   heuristic_smallest_domain,
 
+  derivation_heuristic,
+
   synth,
   SynthResult,
   optimal_program,
   suboptimal_program,
 
+  FixedShapedIterator,
+  UniformIterator,
+  next_solution!,
+
   TopDownIterator,
+  RandomIterator,
   BFSIterator,
   DFSIterator,
   MLFSIterator,

src/fixed_shaped_iterator.jl

@@ -0,0 +1,107 @@
+Base.@doc """
+    @programiterator FixedShapedIterator()
+
+Enumerates all programs that extend from the provided fixed shaped tree.
+The [Solver](@ref) is required to be in a state without any [Hole](@ref)s.
+
+!!! warning: this iterator is used as a baseline for the constraint propagation thesis. After the thesis, this iterator can (and should) be deleted.
+""" FixedShapedIterator
+@programiterator FixedShapedIterator()
+
+"""
+    priority_function(::FixedShapedIterator, g::AbstractGrammar, tree::AbstractRuleNode, parent_value::Union{Real, Tuple{Vararg{Real}}})
+
+Assigns a priority value to a `tree` that needs to be considered later in the search. Trees with the lowest priority value are considered first.
+"""
+function priority_function(
+    ::FixedShapedIterator, 
+    g::AbstractGrammar, 
+    tree::AbstractRuleNode, 
+    parent_value::Union{Real, Tuple{Vararg{Real}}}
+)
+    parent_value + 1;
+end
+
+
+"""
+    hole_heuristic(::FixedShapedIterator, node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
+
+Defines a heuristic over fixed shaped holes. Returns a [`HoleReference`](@ref) once a hole is found.
+"""
+function hole_heuristic(::FixedShapedIterator, node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
+    return heuristic_leftmost_fixed_shaped_hole(node, max_depth);
+end
+
+"""
+    Base.iterate(iter::FixedShapedIterator)
+
+Describes the iteration for a given [`TopDownIterator`](@ref) over the grammar. The iteration constructs a [`PriorityQueue`](@ref) first and then prunes it propagating the active constraints. Recursively returns the result for the priority queue.
+"""
+function Base.iterate(iter::FixedShapedIterator)
+    # Priority queue with number of nodes in the program
+    pq :: PriorityQueue{SolverState, Union{Real, Tuple{Vararg{Real}}}} = PriorityQueue()
+
+    solver = iter.solver
+    @assert !contains_nonuniform_hole(get_tree(iter.solver)) "A FixedShapedIterator cannot iterate partial programs with Holes"
+
+    if isfeasible(solver)
+        enqueue!(pq, get_state(solver), priority_function(iter, get_grammar(solver), get_tree(solver), 0))
+    end
+    return _find_next_complete_tree(solver, pq, iter)
+end
+
+
+"""
+    Base.iterate(iter::FixedShapedIterator, pq::DataStructures.PriorityQueue)
+
+Describes the iteration for a given [`TopDownIterator`](@ref) and a [`PriorityQueue`](@ref) over the grammar without enqueueing new items to the priority queue. Recursively returns the result for the priority queue.
+"""
+function Base.iterate(iter::FixedShapedIterator, pq::DataStructures.PriorityQueue)
+    return _find_next_complete_tree(iter.solver, pq, iter)
+end
+
+"""
+    _find_next_complete_tree(solver::Solver, pq::PriorityQueue, iter::FixedShapedIterator)::Union{Tuple{RuleNode, PriorityQueue}, Nothing}
+
+Takes a priority queue and returns the smallest AST from the grammar it can obtain from the queue or by (repeatedly) expanding trees that are in the queue.
+Returns `nothing` if there are no trees left within the depth limit.
+"""
+function _find_next_complete_tree(
+    solver::Solver, 
+    pq::PriorityQueue,
+    iter::FixedShapedIterator
+)::Union{Tuple{RuleNode, PriorityQueue}, Nothing}
+    while length(pq) ≠ 0
+        (state, priority_value) = dequeue_pair!(pq)
+        load_state!(solver, state)
+
+        hole_res = hole_heuristic(iter, get_tree(solver), typemax(Int))
+        if hole_res ≡ already_complete
+            #the tree is complete
+            return (get_tree(solver), pq)
+        elseif hole_res ≡ limit_reached
+            # The maximum depth is reached
+            continue
+        elseif hole_res isa HoleReference
+            # UniformHole was found
+            (; hole, path) = hole_res
+
+            rules = findall(hole.domain)
+            number_of_rules = length(rules)
+            for (i, rule_index) ∈ enumerate(findall(hole.domain))
+                if i < number_of_rules
+                    state = save_state!(solver)
+                end
+                @assert isfeasible(solver) "Attempting to expand an infeasible tree: $(get_tree(solver))"
+                remove_all_but!(solver, path, rule_index)
+                if isfeasible(solver)
+                    enqueue!(pq, get_state(solver), priority_function(iter, get_grammar(solver), get_tree(solver), priority_value))
+                end
+                if i < number_of_rules
+                    load_state!(solver, state)
+                end
+            end
+        end
+    end
+    return nothing
+end

src/genetic_search_iterator.jl

@@ -113,9 +113,10 @@ Returns the best program within the population with respect to the fitness funct
 function get_best_program(population::Array{RuleNode}, iter::GeneticSearchIterator)::RuleNode
     best_program = nothing
     best_fitness = 0
+    grammar = get_grammar(iter.solver)
     for index ∈ eachindex(population)
         chromosome = population[index]
-        zipped_outputs = zip([example.out for example in iter.spec], execute_on_input(iter.grammar, chromosome, [example.in for example in iter.spec]))
+        zipped_outputs = zip([example.out for example in iter.spec], execute_on_input(grammar, chromosome, [example.in for example in iter.spec]))
         fitness_value = fitness(iter, chromosome, collect(zipped_outputs))
         if isnothing(best_program) 
             best_fitness = fitness_value
@@ -137,13 +138,14 @@ Iterates the search space using a genetic algorithm. First generates a populatio
 """
 function Base.iterate(iter::GeneticSearchIterator)
     validate_iterator(iter)
-    grammar = iter.grammar
+    grammar = get_grammar(iter.solver)
 
     population = Vector{RuleNode}(undef,iter.population_size)
 
+    start_symbol = get_starting_symbol(iter.solver)
     for i in 1:iter.population_size
         # sample a random nodes using start symbol and grammar
-        population[i] = rand(RuleNode, grammar, iter.sym, iter.maximum_initial_population_depth)
+        population[i] = rand(RuleNode, grammar, start_symbol, iter.maximum_initial_population_depth)
     end 
     best_program = get_best_program(population, iter)
     return (best_program, GeneticIteratorState(population))
@@ -160,7 +162,7 @@ function Base.iterate(iter::GeneticSearchIterator, current_state::GeneticIterato
     current_population = current_state.population
 
     # Calculate fitness
-    zipped_outputs(chromosome) = zip([example.out for example in iter.spec], execute_on_input(iter.grammar, chromosome, [example.in for example in iter.spec]))
+    zipped_outputs(chromosome) = zip([example.out for example in iter.spec], execute_on_input(get_grammar(iter.solver), chromosome, [example.in for example in iter.spec]))
     fitness_array = [fitness(iter, chromosome, collect(zipped_outputs(chromosome))) for chromosome in current_population]
 
     new_population = Vector{RuleNode}(undef,iter.population_size)
@@ -187,7 +189,7 @@ function Base.iterate(iter::GeneticSearchIterator, current_state::GeneticIterato
     for chromosome in new_population
         random_number = rand()
         if random_number < iter.mutation_probability
-            mutate!(iter, chromosome, iter.grammar)
+            mutate!(iter, chromosome, get_grammar(iter.solver))
         end
     end
 

src/heuristics.jl

@@ -1,13 +1,41 @@
 using Random
 
+"""
+    heuristic_leftmost_fixed_shaped_hole(node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
+    
+Defines a heuristic over [FixedShapeHole](@ref)s, where the left-most hole always gets considered first. Returns a [`HoleReference`](@ref) once a hole is found. This is the default option for enumerators.
+"""
+function heuristic_leftmost_fixed_shaped_hole(node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
+    function leftmost(node::AbstractRuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
+        if max_depth == 0 return limit_reached end
+
+        for (i, child) in enumerate(node.children)
+            new_path = push!(copy(path), i)
+            hole_res = leftmost(child, max_depth-1, new_path)
+            if (hole_res == limit_reached) || (hole_res isa HoleReference)
+                return hole_res
+            end
+        end
+
+        return already_complete
+    end
+
+    function leftmost(hole::UniformHole, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
+        if max_depth == 0 return limit_reached end
+        return HoleReference(hole, path)
+    end
+
+    return leftmost(node, max_depth, Vector{Int}())
+end
+
 
 """
     heuristic_leftmost(node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
     
 Defines a heuristic over holes, where the left-most hole always gets considered first. Returns a [`HoleReference`](@ref) once a hole is found. This is the default option for enumerators.
 """
 function heuristic_leftmost(node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
-    function leftmost(node::RuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
+    function leftmost(node::AbstractRuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
         if max_depth == 0 return limit_reached end
 
         for (i, child) in enumerate(node.children)
@@ -35,7 +63,7 @@ end
 Defines a heuristic over holes, where the right-most hole always gets considered first. Returns a [`HoleReference`](@ref) once a hole is found. 
 """
 function heuristic_rightmost(node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
-    function rightmost(node::RuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
+    function rightmost(node::AbstractRuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
         if max_depth == 0 return limit_reached end
 
         for (i, child) in Iterators.reverse(enumerate(node.children))
@@ -64,7 +92,7 @@ end
 Defines a heuristic over holes, where random holes get chosen randomly using random exploration. Returns a [`HoleReference`](@ref) once a hole is found.
 """
 function heuristic_random(node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
-    function random(node::RuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
+    function random(node::AbstractRuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
         if max_depth == 0 return limit_reached end
 
         for (i, child) in shuffle(collect(enumerate(node.children)))
@@ -92,7 +120,7 @@ end
 Defines a heuristic over all available holes in the unfinished AST, by considering the size of their respective domains. A domain here describes the number of possible derivations with respect to the constraints. Returns a [`HoleReference`](@ref) once a hole is found. 
 """
 function heuristic_smallest_domain(node::AbstractRuleNode, max_depth::Int)::Union{ExpandFailureReason, HoleReference}
-    function smallest_domain(node::RuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
+    function smallest_domain(node::AbstractRuleNode, max_depth::Int, path::Vector{Int})::Union{ExpandFailureReason, HoleReference}
         if max_depth == 0 return limit_reached end
 
         smallest_size::Int = typemax(Int)