2. [Decision Trees] In class, we discussed early stopping of generating splits and post- pruning. Here,...

Transcribed Image Text:

2. [Decision Trees] In class, we discussed early stopping of generating splits and post- pruning. Here, we consider how they interact. 1 a. (4 pts) Naïvely, one might expect that if we are using post-pruning on a validation set, then there is never any benefit to using early stopping conditions (e.g., maximum depth to split). Explain why this is not the case. b. (4 pts) Suppose we are training a decision tree with both early-stopping conditions and post-pruning. For each of the following, indicate whether it increases bias or variance: (i) increase the maximum depth of the decision tree, (ii) increase the minimum number of samples needed to split, (iii) disable post-pruning, and (iv) assuming we are using a feature map, add more features to the feature map. 2. [Decision Trees] In class, we discussed early stopping of generating splits and post- pruning. Here, we consider how they interact. 1 a. (4 pts) Naïvely, one might expect that if we are using post-pruning on a validation set, then there is never any benefit to using early stopping conditions (e.g., maximum depth to split). Explain why this is not the case. b. (4 pts) Suppose we are training a decision tree with both early-stopping conditions and post-pruning. For each of the following, indicate whether it increases bias or variance: (i) increase the maximum depth of the decision tree, (ii) increase the minimum number of samples needed to split, (iii) disable post-pruning, and (iv) assuming we are using a feature map, add more features to the feature map.

Answer rating: 100% (QA)

Lets address each part of the question one by one Part a Early stopping and postpruning are both techniques used to prevent overfitting in decision trees but they approach the problem at different tim... View the full answer