`init_range_low` is not adhered to in recent release

[cathaloruaidh]

Hi,

I re-installed the most recent release of PyGAD via conda this morning (3.5.0-pyhd8ed1ab_0), and ran some code I had been working on last week. The code takes 20 genes with values of 0 or 1. The fitness function sums the even-numbered genes and subtracts the odd-numbered genes. When I had the previous release installed (3.4.0-pyhd8ed1ab_0), the code gave the correct results, something like:

Parameters of the best solution : [1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0]
Fitness value of the best solution = 10
Predicted output based on the best solution : 10

However, with the new release, I'm getting values for the genes less than init_range_low, for example:

Parameters of the best solution : [  1 -14   1 -12   1  -9   1 -14   1  -9   1 -11   1 -11   1  -8   1 -12  1 -12]
Fitness value of the best solution = 122
Predicted output based on the best solution : 122

When I downgraded to the previous release, I get the correct results as before. The code is as follows:

import pygad
import numpy 
import scipy

def fitness_func(ga_instance, solution, solution_idx):
	fitness = numpy.sum(solution[0::2]) - numpy.sum(solution[1::2])
	return fitness

fitness_function = fitness_func
num_generations = 50
num_parents_mating = 10

sol_per_pop = 100
num_genes = 20

gene_type=int
init_range_low = 0
init_range_high = 2

parent_selection_type = "sss"
keep_parents = 10

crossover_type = "scattered"
mutation_type = "random"
mutation_percent_genes = 100 / num_genes

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, 
gene_type=gene_type,
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)

ga_instance.run()

solution, solution_fitness, solution_idx = ga_instance.best_solution()
print("Parameters of the best solution : {solution}".format(solution=solution))
print("Fitness value of the best solution = {solution_fitness}".format(solution_fitness=solution_fitness))

prediction = numpy.sum(solution[0::2]) - numpy.sum(solution[1::2])
print("Predicted output based on the best solution : {prediction}".format(prediction=prediction))

[ahmedfgad]

In version 3.4.0, the mutation was applied by:

1. Generating a random mutation value.
2. Changing the data type of the random value and rounding it if necessary.
3. Summing the gene value to the random value.
4. Changing the data type again.

For example:

1. The random value is sampled be default from the -1 to 1 range.
2. If the random value is 0.5 and you used the int data type, then it changes to 0.
3. The new random value 0 is added to the gene value. If the gene value is 0 or 1, then it keeps the same value.

Note that the values are sampled from the -1 to 1 range. It might happen that the random value is -1. In this case, -1 is added to 0 and the new gene value will be -1 which is outside the init_range_low to init_range_high range. You were just lucky.

In version 3.5.0, the behaviour changed to:

1. Generate a random mutation value.
2. Summing the gene value to the random value.
3. Changing the data type of the random value and rounding it if necessary.

For example:

1. The random value could be -1.
2. The random value -1 is added to the gene value (e.g. 0) and the result is -1.
3. The data type of the new gene value -1 is changed. The new value keeps -1.

If the gene value is -1 and you again sampled a random value of -1, then the new gene value becomes -2. The process repeats to increase the magnitude of the gene value until it reaches the values you found such as -14.

This is the new working code. Instead of using init_range_low to init_range_high, just use the gene_space parameter.

import pygad
import numpy
import scipy

def fitness_func(ga_instance, solution, solution_idx):
	fitness = numpy.sum(solution[0::2]) - numpy.sum(solution[1::2])
	return fitness

fitness_function = fitness_func
num_generations = 2
num_parents_mating = 10

sol_per_pop = 100
num_genes = 20

gene_type=int
# init_range_low = 0
# init_range_high = 2
gene_space = [0, 1]

parent_selection_type = "sss"
keep_parents = 10

crossover_type = "scattered"
mutation_type = "random"
mutation_percent_genes = 100 / num_genes

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,
gene_type=gene_type,
# init_range_low=init_range_low,
# init_range_high=init_range_high,
gene_space=gene_space,
parent_selection_type=parent_selection_type,
keep_parents=keep_parents,
crossover_type=crossover_type,
mutation_type=mutation_type,
mutation_percent_genes=mutation_percent_genes)

ga_instance.run()

solution, solution_fitness, solution_idx = ga_instance.best_solution()
print("Parameters of the best solution : {solution}".format(solution=solution))
print("Fitness value of the best solution = {solution_fitness}".format(solution_fitness=solution_fitness))

prediction = numpy.sum(solution[0::2]) - numpy.sum(solution[1::2])
print("Predicted output based on the best solution : {prediction}".format(prediction=prediction))

[cathaloruaidh]

Hi @ahmedfgad,

That's great, thanks for the clarification!

Cathal