Merge pull request Wandalen#1116 from YuliaProkopovych/opt-params-cac…

…hing

NOT READY: Caching optimal parameters search results

Cargo.toml

@@ -416,3 +416,6 @@ version = "~0.2.0"
 path = "module/test/c"
 default-features = true
 
+[patch.crates-io]
+pathfinder_geometry = { git = "https://github.com/servo/pathfinder.git" }
+pathfinder_simd = { git = "https://github.com/servo/pathfinder.git" }

module/move/optimization_tools/Cargo.toml

@@ -57,6 +57,8 @@ piston_window = { version = "0.120.0", optional = true }
 exmex = { version = "0.18.0", features = [ "partial" ], optional = true }
 rayon = "1.8.0"
 thiserror = "1.0.56"
+rkyv = { version = "0.7.44", features = [ "validation" ] }
+ordered-float = "4.2.0"
 
 [dev-dependencies]
 test_tools = { workspace = true }

module/move/optimization_tools/src/hybrid_optimizer/gen_alg.rs

@@ -108,17 +108,24 @@ pub trait InitialProblem
 }
 
 /// Indicates state of population proportions with no percentage for elites selection set.
+#[ derive( Debug ) ]
 pub struct NoElites;
 /// Indicates state of population proportions with no percentage for mutating population set.
+#[ derive( Debug ) ]
 pub struct NoMutations;
 /// Indicates state of population proportions with no percentage for crossover set.
+#[ derive( Debug ) ]
 pub struct NoCrossover;
 
 /// Proportion of population modifications with crossover, mutations and elites cloning.
+#[ derive( Debug ) ]
 pub struct PopulationModificationProportions< E, M, C >
 {
+  /// Percent of most fit individuals cloned to next population. 
   elite_selection_rate : E,
+  /// Percent of individuals mutated in new population.
   mutation_rate : M,
+  /// Percent of individuals in new population created by crossover of two selected parents.
   crossover_rate : C,
 }
 

module/move/optimization_tools/src/hybrid_optimizer/mod.rs

@@ -34,6 +34,7 @@ pub enum Reason
   DynastiesLimit,
 }
 
+/// Configuration for Hybrid Optimizer.
 #[ derive( Debug ) ]
 pub struct Config
 {
@@ -92,6 +93,7 @@ impl Default for Config
   }
 }
 
+/// Specific optimization problem for Hybrid Optimizer.
 #[ derive( Debug ) ]
 pub struct Problem< S : InitialProblem, C, M >
 {
@@ -113,6 +115,7 @@ pub struct Problem< S : InitialProblem, C, M >
 
 impl< S : InitialProblem, C, M > Problem< S, C, M >
 {
+  /// Create new instance of optimization problem for Hybrid Optimizer.
   pub fn new( initial : S, crossover_operator : C, mutation_operator : M ) -> Self
   where TournamentSelection : SelectionOperator< < S as InitialProblem >::Person >
   {
@@ -142,9 +145,10 @@ impl< S : InitialProblem, C, M > Problem< S, C, M >
 #[ derive( Debug ) ]
 pub struct HybridOptimizer< S : InitialProblem, C, M >
 {
-
+  /// Configuration of Hybrid Optimizer.
   config : Config,
 
+  /// Specific optimization problem.
   problem : Problem< S, C, M >,
 }
 
@@ -488,6 +492,7 @@ where M : MutationOperator::< Person = < S as InitialProblem >::Person > + Sync,
 
 }
 
+/// Starting parameters for optimal parameters search for hybrid optimization configuration.
 pub fn starting_params_for_hybrid() -> Result< OptimalProblem< RangeInclusive< f64 > >, optimal_params_search::Error >
 {
   let opt_problem = OptimalProblem::new()
@@ -503,6 +508,7 @@ pub fn starting_params_for_hybrid() -> Result< OptimalProblem< RangeInclusive< f
   Ok( opt_problem )
 }
 
+/// Starting parameters for optimal parameters search for SA optimization configuration.
 pub fn starting_params_for_sa() -> Result< OptimalProblem< RangeInclusive< f64 > >, optimal_params_search::Error >
 {
   let opt_problem = OptimalProblem::new()
@@ -518,6 +524,7 @@ pub fn starting_params_for_sa() -> Result< OptimalProblem< RangeInclusive< f64 >
   Ok( opt_problem )
 }
 
+/// Starting parameters for optimal parameters search for GA optimization configuration.
 pub fn starting_params_for_ga() -> Result< OptimalProblem< RangeInclusive< f64 > >, optimal_params_search::Error >
 {
   let opt_problem = OptimalProblem::new()

module/move/optimization_tools/src/optimal_params_search/mod.rs

@@ -1,13 +1,15 @@
 //! Funcions for calculation optimal config parameters.
 //! 
-
+pub mod results_serialize;
 pub mod nelder_mead;
 pub mod sim_annealing;
 use std::ops::RangeBounds;
 use iter_tools::Itertools;
 
 use crate::hybrid_optimizer::*;
 
+use self::results_serialize::read_results;
+
 /// Level of difficulty of sudoku board.
 #[ derive( Debug, Clone, Copy, PartialEq, Eq, Hash ) ]
 pub enum Level
@@ -31,12 +33,18 @@ impl Level {
   }
 }
 
+/// 
 #[ derive( Debug, Clone ) ]
 pub struct OptimalParamsConfig
 {
-  improvement_threshold : f64,
-  max_no_improvement_steps : usize,
-  max_iterations : usize,
+  /// Minimal value detected as improvement in objective function result.
+  pub improvement_threshold : f64,
+
+  /// Max amount of steps performed without detected improvement, termination condition.
+  pub max_no_improvement_steps : usize,
+
+  /// Limit of total iterations of optimization process, termination condition.
+  pub max_iterations : usize,
 }
 
 impl Default for OptimalParamsConfig
@@ -47,22 +55,31 @@ impl Default for OptimalParamsConfig
     {
       improvement_threshold : 0.005,
       max_no_improvement_steps : 10,
-      max_iterations : 100,
+      max_iterations : 10,
     }
   }
 } 
 
+/// Problem for optimal parameters search using Nelder-Mead algorithm.
 #[ derive( Debug, Clone ) ]
 pub struct OptimalProblem< R : RangeBounds< f64 > >
 {
+  /// Containes names of parameters if provided.
   pub params_names : Vec< Option< String > >,
+
+  /// Contains bounds for parameters, may be unbounded or bounded on one side.
   pub bounds : Vec< Option< R > >,
+
+  /// Starting point coordinates for optimization process.
   pub starting_point : Vec< Option< f64 > >,
+
+  /// Size of initial simplex for optimization.
   pub simplex_size : Vec< Option< f64 > >,
 }
 
 impl< 'a, R : RangeBounds< f64 > > OptimalProblem< R >
 {
+  /// Create new instance for optimization problem
   pub fn new() -> Self
   {
     Self
@@ -74,6 +91,7 @@ impl< 'a, R : RangeBounds< f64 > > OptimalProblem< R >
     }
   }
 
+  /// Add parameter to optimal parameters search problem.
   pub fn add
   ( 
     mut self,
@@ -129,24 +147,24 @@ where  R : RangeBounds< f64 > + Sync,
   let ga_crossover_operator = hybrid_problem.ga_crossover_operator.clone();
   let mutation_operator = hybrid_problem.mutation_operator.clone();
 
-  let objective_function = | case : nelder_mead::Point |
+  let objective_function = | case : &nelder_mead::Point |
   {
     log::info!
     (
       "temp_decrease_coefficient : {:.4?}, max_mutations_per_dynasty: {}, mutation_rate: {:.2}, crossover_rate: {:.2};",
-      case.coords[ 0 ], case.coords[ 1 ] as usize, case.coords[ 2 ], case.coords[ 3 ]
+      case.coords[ 0 ], case.coords[ 1 ].into_inner() as usize, case.coords[ 2 ], case.coords[ 3 ]
     );
 
     log::info!
     (
       "max_stale_iterations : {:?}, population_size: {}, dynasties_limit: {};",
-      case.coords[ 4 ] as usize, case.coords[ 5 ] as usize, case.coords[ 6 ] as usize
+      case.coords[ 4 ].into_inner() as usize, case.coords[ 5 ].into_inner() as usize, case.coords[ 6 ].into_inner() as usize
     );
 
     let temp_schedule = LinearTempSchedule
     {
       constant : 0.0.into(),
-      coefficient : case.coords[ 0 ].into(),
+      coefficient : case.coords[ 0 ].into_inner().into(),
       reset_increase_value : 1.0.into(),
     };
 
@@ -160,16 +178,16 @@ where  R : RangeBounds< f64 > + Sync,
     };
 
     let props = crate::hybrid_optimizer::PopulationModificationProportions::new()
-    .set_crossover_rate( case.coords[ 3 ] )
-    .set_mutation_rate( case.coords[ 2 ] )
+    .set_crossover_rate( case.coords[ 3 ].into_inner() )
+    .set_mutation_rate( case.coords[ 2 ].into_inner() )
     ;
 
     let optimizer = HybridOptimizer::new( Config::default(), h_problem )
-    .set_sa_max_mutations_per_dynasty( case.coords[ 1 ] as usize )
+    .set_sa_max_mutations_per_dynasty( case.coords[ 1 ].into_inner() as usize )
     .set_population_proportions( props )
-    .set_max_stale_iterations( case.coords[ 4 ] as usize )
-    .set_population_size( case.coords[ 5 ] as usize )
-    .set_dynasties_limit( case.coords[ 6 ] as usize )
+    .set_max_stale_iterations( case.coords[ 4 ].into_inner() as usize )
+    .set_population_size( case.coords[ 5 ].into_inner() as usize )
+    .set_dynasties_limit( case.coords[ 6 ].into_inner() as usize )
     ;
     let ( _reason, _solution ) = optimizer.optimize();
   };
@@ -181,10 +199,11 @@ where  R : RangeBounds< f64 > + Sync,
   res
 }
 
+/// Wrapper for optimizing objective function by execution time instead of value.
 pub fn optimize_by_time< F, R >( config : OptimalParamsConfig, problem : OptimalProblem< R >, objective_function : F ) -> Result< nelder_mead::Solution, nelder_mead::Error >
-where F : Fn( nelder_mead::Point ) + Sync, R : RangeBounds< f64 > + Sync
+where F : Fn( &nelder_mead::Point ) + Sync, R : RangeBounds< f64 > + Sync
 {
-  let objective_function = | case : nelder_mead::Point |
+  let objective_function = | case : &nelder_mead::Point |
   {
 
     let now = std::time::Instant::now();
@@ -199,39 +218,48 @@ where F : Fn( nelder_mead::Point ) + Sync, R : RangeBounds< f64 > + Sync
     elapsed.as_secs_f64()
   }; 
 
-  let mut bounds = Vec::new();
-  for bound in problem.bounds
-  {
-    if let Some( bound ) = bound
-    {
-      bounds.push( bound );
-    }
-  }
+  // let mut bounds = Vec::new();
+  // for bound in problem.bounds
+  // {
+  //   if let Some( bound ) = bound
+  //   {
+  //     bounds.push( bound );
+  //   }
+  // }
 
-  let mut optimizer = sim_annealing::Optimizer
-  {
-    bounds : bounds,
-    objective_function : objective_function,
-    max_iterations : 50,
-  };
-  // let mut optimizer = nelder_mead::Optimizer::new( objective_function );
-  // optimizer.bounds = problem.bounds;
-  // optimizer.set_starting_point( problem.starting_point.clone() );
-  // optimizer.set_simplex_size( problem.simplex_size );
+  // let optimizer = sim_annealing::Optimizer
+  // {
+  //   bounds : bounds,
+  //   objective_function : objective_function,
+  //   max_iterations : 50,
+  // };
+  let mut optimizer = nelder_mead::Optimizer::new( objective_function );
+  optimizer.bounds = problem.bounds;
+  optimizer.set_starting_point( problem.starting_point.clone() );
+  optimizer.set_simplex_size( problem.simplex_size );
 
-  // optimizer.improvement_threshold = config.improvement_threshold;
-  // optimizer.max_iterations = config.max_iterations;
-  // optimizer.max_no_improvement_steps = config.max_no_improvement_steps;
+  optimizer.improvement_threshold = config.improvement_threshold;
+  optimizer.max_iterations = config.max_iterations;
+  optimizer.max_no_improvement_steps = config.max_no_improvement_steps;
 
-  optimizer.optimize()
+  let calculated_points = read_results();
+  if let Ok( calculated_points ) = calculated_points
+  {
+    optimizer.set_calculated_results( calculated_points );
+  }
+
+  optimizer.optimize_from_random_points()
 }
 
 /// Possible error when building NMOptimizer.
 #[ derive( thiserror::Error, Debug ) ]
-pub enum Error {
+pub enum Error 
+{
+  /// Error for parameters with duplicate names.
   #[ error( "parameter with similar name exists" ) ]
   NameError,
 
+  /// Error for value located out of its bounds.
   #[ error( "starting value is out of bounds" ) ]
   OutOfBoundsError,
 }