This is an attempt at a library of search algorithm with a straightforward API and no unnecessary restrictions (e.g. in the types of states and costs). The user should not be required to subclass any internal type of object. However, object types may be required to have certain properties (e.g. states are in some algorithms required to be hashable; cost objects can be of any type but are usually required to support comparison operators and sometimes addition). Callback function parameters are the way the algorithms get the things they need (e.g. solutions in the neighborhood, an action's cost, a solution's fitness, etc.).
Current search algorithms:
- A* (
from searchlib import astar) - Simulated Annealing (
from searchlib import simulated_annealing) - Tabu Search (
from searchlib import tabu) - Genetic Algorithm (
from searchlib import genetic_algorithm)
- Contributions welcome!
- One file per search algorithm
- Callback functions > forcing the user to subclass certain classes
- Building:
python3 setup.py build_ext --inplace pip3 install git+git://github.com/ctittel/searchlib(orpip3 install .if you have the repository locally and arecd'd into it)
- Add more algorithms: ACO, Greate Deluge
- Add test cases and examples for all algorithms
- Optimization as much as possible (currently with Cython, can it be improved without introducing any more restrictions for the user?)
- "Essentials of Metaheuristics" by Sean Luke (Link: http://cs.gmu.edu/%7Esean/book/metaheuristics/)