How to define a problem and solve it using a metaheuristic (Solving Use Case)

A solver have to make the following steps to solve a optimization problem (typically a hyperparameter optimization problem) using a MetaGen:

Define the problem space using the Domain class.
Implement the fitness function using the Solution class.
Choose a metaheuristic to solve the problem (can be a MetaGen’s implemented one or a custom one), instantiate it, and run it.

For example, to solve the following simple problem \(x + 5\) where \(x \in \mathbb{I},\ -5 \leq x \leq 10\), you can use the Random Search algorithm by coding the following Python script:

from metagen.framework import Domain, Solution
from metagen.heuristics import RandomSearch

# Step 1: Define the problem space
domain = Domain()
domain.define_integer("x", -5, 10)

# Step 2: Implement the fitness function
def fitness(solution: Solution) -> float:
    x = solution["x"]
    return x+5

# Step 3: Choose a metaheuristic and run it
best_solution: Solution = RandomSearch(domain, fitness).run()

Or, to solve another simple problem like \(x^2\) where \(x \in \mathbb{R},\ 0 \leq x, y \leq 1\), you can use the Random Search algorithm by coding the following Python script:

from metagen.framework import Domain, Solution
from metagen.heuristics import RandomSearch

# Step 1: Define the problem space
domain = Domain()
domain.define_real("x", 0, 1)

# Step 2: Implement the fitness function
def fitness(solution: Solution) -> float:
    x = solution["x"]
    return x**2

# Step 3: Choose a metaheuristic and run it
best_solution: Solution = RandomSearch(domain, fitness).run()

Both simple examples can be executed by these Google Colab Notebooks:

x+5
x²

Next, we will show how to use MetaGen to solve two hyperparameter optimization problems to tune a Random Forest Classifier using the scikit-learn library and a Deep Neural Network using the TensorFlow library.