Hands-On Genetic Algorithms with Python

Solution representation

Regarding our existing network architecture configuration of four floats of the form [n1, n2, n3, n4], we can add the following four hyperparameters:

• activation can have one of three values: 'tanh', 'relu', or 'logistic'. This can be achieved by representing it as a float number in the range of [0, 2.99]. To transform the float value into one of the aforementioned values, we need to apply the floor() function to it, which will yield either 0, 1, or 2. We then replace a value of 0 with tanh, a value of 1 with relu, and a value of 2 with logistic.
• solver can have one of three values: 'sgd', 'adam', or 'lbfgs'. Just like the activation parameter, it can be represented using a float number in the range of [0, 2.99]. ...