simple-genetic-algorithm 0.1.8

Genetic Algorithm Framework

Genetic-Algorithm

Implementation of Genetic-Algorithm for solution finding (optimization)

Easy to use GA implementation. With parallel computing and info-prints. Simple and flexible for your optimal solution finding.

Usage

1. Get the code

   • Download the project and add to python module search path in your code

     import sys
     sys.path.insert(0, '../path_to_GA_py_dir')
     # ./ => this folder
     # ../ => supfolder
     

   • Or pip install it (easier)

     pip install Simple-Genetic-Algorithm
     

2. Import the class and helper function

   from genetic_algorithm import GA, get_random
   

3. Create 2 functions and parameters

   class Example_GA(GA):
   
       def calculate_fitness(self, kwargs, params):
           # return here the fitness (how good the solution is)
           # as bigger as better!
           # hint: if you have a loss, you should propably just return -1*loss
           # example for sklearn model:
            model = RandomForestRegressor(n_estimators=params["n_estimators"], ...)
     		 model = model.fit(kwargs["X_train"], kwargs["y_train"])
           # predict
           y_pred = model.predict(kwargs["X_test"])
           # calc mean absolute error loss
           y_true = np.array(kwargs["y_test"])
           y_pred = np.array(y_pred)
           return - np.mean(np.abs(y_true - y_pred))
   
       def get_random_value(self, param_key):
           if param_key == "name_of_parameter_1":
               return get_random(10, 1000)
           elif param_key == "name_of_parameter_2":
               return get_random(["something_1", "something_2", "something_3"])
           elif param_key == "name_of_parameter_3":
               return get_random(0.0, 1.0)
           
           ...
   
   parameters = ["n_estimators", "criterion", "max_depth", "max_features", "bootstrap"]
   

4. Create and run genetic algorithm and pass the input, which will be used in the calculate_fitness function (in kwargs variable)

   optimizer = Example_GA(generations=10, population_size=15, mutation_rate=0.3, list_of_params=parameters)
   optimizer.optimize(X_train=X_train, y_train=y_train, X_test=X_dev, y_test=y_dev)
   

Short explanation:
The kwargs are the inputs of optimize-method. These are the values which are needed to calculate the fitness. Maybe you can calculate the fitness without them, depending on what you are optimizing.
The list of parameters are the gene/the solution, so the parameters which are changed and optimized.
The get_random_value method return a random value for a given parameter, so that the solutions can be initialized and mutated.

For more examples, see:

• Regression with RandomForrestRegressor
• Knapsack problem

License

Feel free to use it. Of course you don't have to name me in your code :)

-> It is a copy-left license

For all details see the license file.