Skip to content

Latest commit

 

History

History
118 lines (81 loc) · 5.01 KB

README.md

File metadata and controls

118 lines (81 loc) · 5.01 KB
Raw

General-Optimization-Solver

Welcome to the README.md file of the General Optimization Solver.

General Optimization Solver is practical framework for Combinatorial Optimization designed to facilitate unified benchmark handling and simplify the integration of custom solvers and their comparison. Prioritizing simplicity and accessibility, it supports Genetic Algorithms and Constraint Programming solvers defined with pymoo and IBM’s DOCplex CP Optimizer, respectively. The framework facilitates handling complex problems such as RCPSP, Job Shop Scheduling problems, and Packing problems. It is not intended to provide state-ofthe-art solutions but rather to offer a reliable and easy-to-use tool for researchers. Tool is based on modular architecture that makes it easy extend it with new benchmarks, problems or solvers.

The project is divided into 3 main folders

  • src constaing modules with the source code of the framework
  • examples containing examples of using the framework
  • raw_data containing raw data of the benchmarks

PROJECT STRUCTURE

Each problem type has its own directory, which contains solver directory with solver modules and problem.py module as well as parser modules, if only one input format is implemented or multiple modules with respective format names (c15.py and mmlib.py for MM-RCPSP)

Base classes are located in common directory

All important classes are imported to the general_optimization_solver.py module, which is the main entry point for the project

Framework API

- load_raw_benchmark(directory_path, solution_path, format, no_instances, force_dump)
  - load raw data from specified files in a specified format

- load_benchmark(directory_path, no_instances)
  - load unified data from json files in specified directory

- load_raw_instance(path, solution_path, format, verbose)

- load_instance(path):

Parsers

Collection of '{benchmark_name}.py' modules which should contain load_{benchmark_name} which load instance file and load_{benchmark_name}_solution which loads instance solution functions.

Instance loader should return json with keys similar to the specific PROBLEM class.

Instance solution loader should return following json {"feasible": None, "optimum": None, "cpu_time": None, "bounds": None} (cpu_time and bounds being optional)

Solvers

Solver class:

- solve(): -- abstract

CPSolver class:

- __init__(TimeLimit, no_workers):  
    - solved  
    - params  

- solve(instance, validate, visualize, force_execution): - CP solver specific (base abstract)
  - so far returns docplex solution and cp variables

- add_run_to_history(instance, sol): - CP Solver specific (implemented in base class)
  - extracts common info from docplex solution and updates the instance run history

GASolver class:

- solve(algorithm, instance, fitness_func, termination, validate, visualize, force_execution): - GA solver specific
  - returns objective value, {other important solution values like start times for rcpsp}, pymoo solution object

- add_run_to_history(instance, objective_value, start_times, is_valid): - GA Solver specific (implemented in base class)
  - extracts common info from pymoo solution and updates the instance run history

PROBLEMS

Benchmark class:

- __init__(name, instances):  
    _name  
    _instances  

- solve(solver, method, solver_config, force_dump)
  - for each instance in benchmark, run solver.solve method 
  - solver config is method (CP/GA) specific

- dump()
  - calls OptimizationProble.dump_json for each instance in benchmark

- generate_solver_comparison_markdown_table(instances_subset, methods_subset):
  - generate markdown table with comparison of solver results for given instances and methods

OptimizationProblem class:

- __init__(benchmark_name, instance_name, _instance_kind, data, solution, run_history):  
  - General config  
    - _benchmark_name  
    - _instance_name  
    - _instance_kind  
    - _data  
    - _solution  
    - _run_history  
  - Problem specific config  
    - ...

- dump():
  - save json to "data/{self._instance_kind}/{self._benchmark_name}/{self._instance_name}.json"
  - dict prescription:
    {
        "benchmark_name": self._benchmark_name,
        "instance_name": self._instance_name,
        "instance_kind": self._instance_kind,

        "data": self._data,
        "reference_solution": self._solution,
        "run_history": self._run_history,
    }
- compare_to_reference(obj_value):
  - print out whether solution is optimal or worse than optimal if optimum is known
  - print out whether solution is worse than upper / lower bound
  - print out if no solution exists

- update_run_history(solution, method, solver_params)
- reset_run_history()

- skip_on_optimal_solution

- visualize(solution, variables) - problem specific
- validate(solution, variables) - problem specific