GeneticAlgorithmPython is an open source Python project for implementing the genetic algorithm based on NumPy. Based on this project, a Python library named PyGAD is available at PyPI: https://pypi.org/project/pygad. To install PyGAD using pip, just issue this command:
pip install pygadThe project has a single module named pygad.py which contains a class named GA. There is a script named example.py in which an example is given for using the pygad.py module.
Simply, to run the genetic algorithm all you need to do is to create an instance of the GA class and pass the appropriate parameters to its constructor. The GA class has all the required parameters and methods for implementing the genetic algorithm.
The documentation starts by discussing the available parameters in addition to the steps of using the library. Note that the PyGAD library is a reflection to the GeneticAlgorithmPython GitHub project. As a result, the documentation can either use project or library interchangeably.
The single module available in the PyGAD library is named pygad.py and contains a class named GA. For creating an instance of this class, there are a number of parameters that allows the user to customize the genetic algorithm. Before running the GA, the parameters must be prepared. The list of all supported parameters is as follows:
num_generations: Number of generations.num_parents_mating: Number of solutions to be selected as parents.fitness_func: Accepts a function that must accept 2 parameters (a single solution and its index in the population) and return the fitness value of the solution. Available starting from PyGAD 1.0.17 until 1.0.20 with a single parameter representing the solution. Changed in PyGAD 2.0.0 and higher to include the second parameter representing the solution index.initial_population: A user-defined initial population. It is useful when the user wants to start the generations with a custom initial population. It defaults toNonewhich means no initial population is specified by the user. In this case, PyGAD creates an initial population using thesol_per_popandnum_genesparameters. An exception is raised if theinitial_populationisNonewhile any of the 2 parameters (sol_per_popornum_genes) is alsoNone. Introduced in PyGAD 2.0.0 and higher.sol_per_pop: Number of solutions (i.e. chromosomes) within the population.num_genes: Number of genes in the solution/chromosome.init_range_low=-4: The lower value of the random range from which the gene values in the initial population are selected.init_range_lowdefaults to-4. Available in PyGAD 1.0.20 and higher.init_range_high=4: The upper value of the random range from which the gene values in the initial population are selected.init_range_highdefaults to+4. Available in PyGAD 1.0.20 and higher.parent_selection_type="sss": The parent selection type. Supported types aresss(for steady state selection),rws(for roulette wheel selection),sus(for stochastic universal selection),rank(for rank selection),random(for random selection), andtournament(for tournament selection).keep_parents=-1: Number of parents to keep in the current population.-1(default) means keep all parents in the next population.0means keep no parents in the next population. A valuegreater than 0means keep the specified number of parents in the next population. Note that the value assigned tokeep_parentscannot be< - 1or greater than the number of solutions within the populationsol_per_pop.K_tournament=3: In case that the parent selection type istournament, theK_tournamentspecifies the number of parents participating in the tournament selection. It defaults to3.crossover_type="single_point": Type of the crossover operation. Supported types aresingle_point(for single point crossover),two_points(for two points crossover), anduniform(for uniform crossover). It defaults tosingle_point.mutation_type="random": Type of the mutation operation. Supported types arerandom(for random mutation),swap(for swap mutation),inversion(for inversion mutation), andscramble(for scramble mutation). It defaults torandom.mutation_percent_genes=10: Percentage of genes to mutate which defaults to10. Out of this percentage, the number of genes to mutate is deduced. This parameter has no action if the parametermutation_num_genesexists.mutation_num_genes=None: Number of genes to mutate which defaults toNonemeaning that no number is specified. If the parametermutation_num_genesexists, then no need for the parametermutation_percent_genes.random_mutation_min_val=-1.0: Forrandommutation, therandom_mutation_min_valparameter specifies the start value of the range from which a random value is selected to be added to the gene. It defaults to-1.random_mutation_max_val=1.0: Forrandommutation, therandom_mutation_max_valparameter specifies the end value of the range from which a random value is selected to be added to the gene. It defaults to+1.callback_generation: If notNone, then it accepts a function to be called after each generation. This function must accept a single parameter representing the instance of the genetic algorithm.
The user doesn't have to specify all of such parameters while creating an instance of the GA class. A very important parameter you must care about is fitness_func which defines the fitness function.
It is OK to set the value of any of the 2 parameters init_range_low and init_range_high to be equal, higher or lower than the other parameter (i.e. init_range_low is not needed to be lower than init_range_high).
All the parameters and functions passed to the GA class constructor are used as attributes and methods in the instances of the GA class. In addition to such attributes, there are other attributes and methods added to the instances of the GA class which are:
generations_completed: Holds the number of the last completed generation.population: A NumPy array holding the initial population.valid_parameters: Set toTruewhen all the parameters passed in theGAclass constructor are valid.run_completed: Set toTrueonly after therun()method completes gracefully.pop_size: The population size.crossover: Refers to the method that applies the crossover operator based on the selected type of crossover in thecrossover_typeproperty.mutation: Refers to the method that applies the mutation operator based on the selected type of mutation in themutation_typeproperty.select_parents: Refers to a method that selects the parents based on the parent selection type specified in theparent_selection_typeattribute.best_solution_fitness: A list holding the fitness value of the best solution for each generation.cal_pop_fitness: A method that calculates the fitness values for all solutions within the population by calling the function passed to thefitness_funcparameter for each solution.
Next, the steps of using the PyGAD library are discussed.
To use PyGAD, here is a summary of the required steps:
- Preparing the
fitness_funcparameter. - Preparing other parameters.
- Example of preparing the parameters.
- Import the
pygad.pymodule. - Create an instance of the
GAclass. - Run the genetic algorithm.
- Plotting Results.
- Saving & Loading the Results.
Let's discuss how to do each of these steps.
Even there are a number of steps in the genetic algorithm pipeline that can work the same regardless of the problem being solved, one critical step is the calculation of the fitness value. There is no unique way of calculating the fitness value and it changes from one problem to another.
On 15 April 2020, a new argument named fitness_func is added to PyGAD 1.0.17 that allows the user to specify a custom function to be used as a fitness function. This function must be a maximization function so that a solution with a high fitness value returned is selected compared to a solution with a low value. Doing that allows the user to freely use the library to solve any problem by passing the appropriate fitness function.
Let's discuss an example:
Given the following function: y = f(w1:w6) = w1x1 + w2x2 + w3x3 + w4x4 + w5x5 + 6wx6 where (x1,x2,x3,x4,x5,x6)=(4,-2,3.5,5,-11,-4.7) and y=44 What are the best values for the 6 weights (w1 to w6)? We are going to use the genetic algorithm to optimize this function.
So, the task is about using the genetic algorithm to find the best values for the 6 weight W1 to W6. Thinking of the problem, it is clear that the best solution is that returning an output that is close to the desired output y=44. So, the fitness function should return a value that gets higher when the solution's output is closer to y=44. Here is a function that does that:
function_inputs = [4,-2,3.5,5,-11,-4.7] # Function inputs.
desired_output = 44 # Function output.
def fitness_func(solution, solution_idx):
output = numpy.sum(solution*function_inputs)
fitness = 1.0 / numpy.abs(output - desired_output)
return fitnessThe function must accept 2 parameters:
- 1D vector representing a single solution. Introduced in PyGAD 1.0.17 and higher.
- Solution index within the population. Introduced in PyGAD 2.0.0 and higher.
By creating this function, you are ready to use the library.
Here is an example for preparing the parameters:
num_generations = 50
num_parents_mating = 4
fitness_function = fitness_func
sol_per_pop = 8
num_genes = len(function_inputs)
init_range_low = -2
init_range_high = 5
parent_selection_type = "sss"
keep_parents = 1
crossover_type = "single_point"
mutation_type = "random"
mutation_percent_genes = 10In PyGAD 2.0.0 and higher, an optional parameter named callback_generation is supported which allow the user to call a function (with a single parameter) after each generation. Here is a simple function that just prints the current generation number and the fitness value of the best solution in the current generation. The generations_completed attribute of the GA class returns the number of the last completed generation.
def callback_gen(ga_instance):
print("Generation : ", ga_instance.generations_completed)
print("Fitness of the best solution :", ga_instance.best_solution()[1])After being defined, the function is assigned to the callback_generation parameter of the GA class constructor. By doing that, the callback_gen() function will be called after each generation.
ga_instance = pygad.GA(...,
callback_generation=callback_gen,
...)After the parameters are prepared, we can import the pygad module and build an instance of the GA class.
The next step is to import the pygad module as follows:
import pygadThis module has a class named GA which holds the implementation of all methods for running the genetic algorithm.
The GA class is instantiated where the previously prepared parameters are fed to its constructor. The constructor is responsible for creating the initial population.
ga_instance = pygad.GA(num_generations=num_generations,
num_parents_mating=num_parents_mating,
fitness_func=fitness_function,
sol_per_pop=sol_per_pop,
num_genes=num_genes,
init_range_low=init_range_low,
init_range_high=init_range_high,
parent_selection_type=parent_selection_type,
keep_parents=keep_parents,
crossover_type=crossover_type,
mutation_type=mutation_type,
mutation_percent_genes=mutation_percent_genes)After an instance of the GA class is created, the next step is to call the run() method as follows:
ga_instance.run()Inside this method, the genetic algorithm evolves over a number of generations by doing the following tasks:
- Calculating the fitness values of the solutions within the current population.
- Select the best solutions as parents in the mating pool.
- Apply the crossover & mutation operation
- Repeat the process for the specified number of generations.
There is a method named plot_result() which creates a figure summarizing how the fitness values of the solutions change with the generations .
ga_instance.plot_result()
After the run() method completes, it is possible to save the current instance of the genetic algorithm to avoid losing the progress made. The save() method is available for that purpose. According to the next code, a file named genetic.pkl will be created and saved in the current directory.
# Saving the GA instance.
filename = 'genetic' # The filename to which the instance is saved. The name is without extension.
ga_instance.save(filename=filename)You can also load the saved model using the load() function and continue using it. For example, you might run the genetic algorithm for a number of generations, save its current state using the save() method, load the model using the load() function, and then call the run() method again.
# Loading the saved GA instance.
loaded_ga_instance = pygad.load(filename=filename)After the instance is loaded, you can use it to run any method or access any property.
print(loaded_ga_instance.best_solution())The library supports different types for selecting the parents and applying the crossover & mutation operators. More features will be added in the future. To ask for a feature, please open an issue in the GitHub project: https://github.com/ahmedfgad/GeneticAlgorithmPython/issues/new
The supported crossover operations at this time are:
- Single point: Implemented using the
single_point_crossover()method. - Two points: Implemented using the
two_points_crossover()method. - Uniform: Implemented using the
uniform_crossover()method.
The supported mutation operations at this time are:
- Random: Implemented using the
random_mutation()method. - Swap: Implemented using the
swap_mutation()method. - Inversion: Implemented using the
inversion_mutation()method. - Scramble: Implemented using the
scramble_mutation()method.
The supported parent selection techniques at this time are:
- Steady state: Implemented using the
steady_state_selection()method. - Roulette wheel: Implemented using the
roulette_wheel_selection()method. - Stochastic universal: Implemented using the
stochastic_universal_selection()method. - Rank: Implemented using the
rank_selection()method. - Random: Implemented using the
random_selection()method. - Tournament: Implemented using the
tournament_selection()method.
More types will be added in the future. You can also ask for supporting more types by opening an issue in the GitHub project associated with the library: https://github.com/ahmedfgad/GeneticAlgorithmPython
PyGAD 1.0.17 (15 April 2020):
- The
GAclass accepts a new argument namedfitness_funcwhich accepts a function to be used for calculating the fitness values for the solutions. This allows the project to be customized to any problem by building the right fitness function.
PyGAD 1.0.20 (4 May 2020):
- The attributes are moved from the class scope to the instance scope.
- Raising a
ValueErrorexception on passing incorrect values to the parameters. - Two new parameters are added (
init_range_lowandinit_range_high) allowing the user to customize the range from which the genes values in the initial population are selected. - The code object
__code__of the passed fitness function is checked to ensure it has the right number of parameters.
PyGAD 2.0.0 (13 May 2020)
- The fitness function accepts a new argument named
sol_idxrepresenting the index of the solution within the population. - A new parameter to the GA constructor named
initial_populationis supported to allow the user to use a custom initial population to be used by the genetic algorithm. If not None, then the passed population will be used. IfNone, then the genetic algorithm will create the initial population using thesol_per_popandnum_genesparameters. - The parameters
sol_per_popandnum_genesare optional and set toNoneby default. - A new parameter named
callback_generationis introduced in the GA class constructor. It accepts a function with a single parameter representing the GA instance. This function called after each generation. This helps the user to do post-processing or debugging operations after each generation.
To start with coding the genetic algorithm, you can check the tutorial titled Genetic Algorithm Implementation in Python available at these links:
-
https://www.linkedin.com/pulse/genetic-algorithm-implementation-python-ahmed-gad
-
https://towardsdatascience.com/genetic-algorithm-implementation-in-python-5ab67bb124a6
This tutorial is prepared based on a previous version of the project but it still a good resource to start with coding the genetic algorithm.
You can also check my book cited as Ahmed Fawzy Gad 'Practical Computer Vision Applications Using Deep Learning with CNNs'. Dec. 2018, Apress, 978-1-4842-4167-7.
Important Note
The library just supports the decimal representation for the chromosome and there is no current support for binary representations. The library is updated and soon support for the binary version will be available.