V ariable selection is often of fundamental interest in many data science applications, providing benefits in prediction error, interpretability, and computation by excluding unnecessary variables.

Proposition 2. Using the same notations as in Proposition 1, we have the following results. Algorithm 2 gives pseudocode for finding the optimal split for a given feature. Output: Split (f, t) that gives the largest risk reduction. Proposition 5. F or the sigmoid loss, we have null R Proposition 4. If a node contains the examples Output: Collection of trained decision trees. Algorithm 5: Find_Split(κ, F, T) Input: κ - node; F - number of attributes; T - number of threshold values per attribute.

We introduce a new family of techniques to post-process ("wrap") a black-box classifier in order to reduce its bias.

We thank the reviewers for giving positive and insightful evaluations of our paper. Specific responses are given below. We will discuss these in our paper. Future work could replace Eureqa inside our framework with more sophisticated SR backends. This Eureqa alternative is optimized for rediscovering existing equations by, e.g., This approach does not seem applicable for discovering new equations so we chose Eureqa.