Case study 2 – using HTDM with a conditional hyperparameter space
Let’s say we are faced with a similar condition as in the previous section but now, we are working with a conditional hyperparameter space, as defined here:
one_hot_max_size = randint(2,15)
iterations = randint(5,200)
If iterations < 50:
depth = randint(3,10)
learning_rate = np.linspace(5e-4,1e-3,10)
l2_leaf_reg = np.linspace(1,15,20)
elif iterations < 100:
depth = randint(3,7)
learning_rate = np.linspace(1e-5,5e-4,10)
l2_leaf_reg = np.linspace(5,20,20)
else:
depth = randint(3,5)
learning_rate = np.linspace(1e-6,5e-5,10)
l2_leaf_reg = np.linspace(5,30,20)
Based on the given case description, we can try to utilize HTDM again to help us choose which hyperparameter tuning method suits the condition the best....