Total Variation Floodgate for Variable Importance Inference in Classification

Authors: Wenshuo Wang, Lucas Janson, Lihua Lei, Aaditya Ramdas

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show the effectiveness of our algorithms with simulations and a case study in conjoint analysis on the US general election.
Researcher Affiliation Academia 1Department of Statistics, Harvard University, Cambridge, MA, USA 2Stanford Graduate School of Business, Stanford University, Stanford, CA, USA 3Departments of Statistics and Machine Learning, Carnegie Mellon University, Pittsburgh, PA, USA.
Pseudocode Yes Algorithm 1 Total variation floodgate. ... Algorithm 2 Cross-validated total variation floodgate. ... Algorithm 3 Hierarchically weighted TV floodgate.
Open Source Code Yes The code to implement ETV floodgate and replicate all experiments is available at https://github.com/ wenshuow/etv_floodgate.
Open Datasets Yes We analyze the election data in Ono & Burden (2019), which is under the CC0 license.
Dataset Splits Yes We set p = 4 or 10, β = (0, 1, 2, 3) for p = 4 and β = (0, 0, 0, 0, 1, 2, 3, 4, 5, 6) for p = 10, n = 100p, and apply 10-fold cross validation in Algorithm 2.
Hardware Specification No The paper mentions that experiments run 'on a single CPU' but provides no further specific details about the CPU model, speed, or other hardware components.
Software Dependencies No The paper mentions using Hier Net models and provides a GitHub link to the code. However, it does not list specific software dependencies (e.g., programming languages, libraries, or frameworks) with their version numbers within the text.
Experiment Setup Yes We set p = 4 or 10... n = 100p, and apply 10-fold cross validation in Algorithm 2. ... We apply Algorithm 2 with k = 10 and J = 100. The classifier family f is chosen to be the model-based f in equation (5), where the models are Hier Net (Bien et al., 2013)... where pθ1(y | x, z) is a Hier Net model with a fixed penalty parameter...