Case study 1 – using HTDM with a CatBoost classifier
Let’s say we are training a classifier based on the marketing campaign data that was introduced in Chapter 7, Hyperparameter Tuning via scikit. Here, we are utilizing CatBoost (see Chapter 11, Understanding Hyperparameters of Popular Algorithms) as the classifier. This is our first time working with the given data. The laptop we are using only has a single-core CPU and the hyperparameter space is defined as follows. Note that we are not working with a conditional hyperparameter space:
iterations
:randint(5,200)
depth
:randint(3,10)
learning_rate
:np.linspace(1e-5,1e-3,20)
l2_leaf_reg
:np.linspace(1,30,30)
one_hot_max_size
:randint(2,15)
Based on the given case description, we can try to utilize HTDM to help us choose which hyperparameter tuning suits the condition the best. First of all, we know that we do not have any prior knowledge or meta-learning results of the good hyperparameter...