Hi everyone!
I’m attending the “Data Scientist: Machine Learning Specialist” path.
Going through the project “Find the Flag!”, about Decision Tree Classification, I arrived at question 11, where it asks to tune the ccp_alpha value in DecisionTreeClassifier, in the same way we did previously for the hyperparameter depth.
However, in depth’s case, the initial range where to loop in was already provided (range(1,21)), while in ccp_alpha it wasn’t.
Looking at the documentation, I understood that this value should be a non-negative float, but I was totally unsure about which initial range to use.
I had the idea to start with a big range (using the same range as for depth), and then depending on the accuracy distribution, narrow it in the region showing the highest accuracy (which was between 0 and 1).
The codecademy solution however sets the initial range with ccp = np.logspace(-3, 0, num=20).
My question is: why did they use a log distribution of values as the initial range for ccp_alpha? How did they decide the boundaries?
And more in general: is there a sort of rule to follow in terms of defining the initial range for ccp_alpha tuning?
Thanks for the help!
Happy coding
Alberto