Table 2. Hyperparameters and their optimisation search space.

Hyperparameter	Distribution	Range
Number of units per hidden layer	Quantized uniform	[10, 200]
Learning rate in RBM unsupervised training	Uniform	[1e-1, 1e-4]
Learning rate of the gradient descent algorithm	Uniform	[1e-1, 1e-4]

Notes: Each search space is composed of the original distribution type and range. These search space values are used in the sampling of hyperparameter values in each optimisation iteration. At the end of the iteration, the distribution is modified according to the classifier performance.