Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Conformal Inference for Online Prediction with Arbitrary Distribution Shifts

Authors: Isaac Gibbs, Emmanuel J. Candès

JMLR 2024 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our techniques on two real-world datasets aimed at predicting stock market volatility and COVID-19 case counts and find that they are robust and adaptive to real-world distribution shifts. Keywords: Conformal inference, online prediction, distribution shift, prediction set, online convex optimization
Researcher Affiliation Academia Isaac Gibbs EMAIL Department of Statistics Stanford University Stanford, CA 94305, USA Emmanuel Cand es EMAIL Departments of Statistics and Mathematics Stanford University Stanford, CA 94305, USA
Pseudocode Yes Algorithm 1: Dt ACI, modified version of Algorithm 1 in Gradu et al. (2023). ... Algorithm 2:
Open Source Code Yes Code for reproducing these results can be found at https://github.com/isgibbs/Dt ACI.
Open Datasets Yes All data is obtained from a public repository made available by the DELPHI group at Carnegie Mellon (Reinhart et al. (2021)).
Dataset Splits Yes For our first real-world data example, we return to a stock market prediction task... At each time step, t, we use the most recent 1250 days of returns {Rs}t 1250 s<t to produce estimates... To compute prediction sets for county i, we define the conformity scores St,i := |d COt,i COt,i|/|COt 7,i COt,i| and counts nt := |{i : County i has available data at time t 1}|, and set ˆCt,i(αt,i) := c : |d COt,i c| |COt 7,i c| Quantile.
Hardware Specification No No specific hardware details (like GPU/CPU models, memory, or cloud instances) are mentioned in the paper.
Software Dependencies No The paper does not provide specific version numbers for software libraries, frameworks, or programming languages used in the experiments.
Experiment Setup Yes In all experiments the set of candidate γ values is taken to be {0.001, 0.002, 0.004, 0.008, 0.0160, 0.032, 0.064, 0.128}. ... In our experiments, we will set η and σ using the choice |I| = 500. ... The hyperparameters to be m = 40, η = q 4.2m , and r = 800000. ... We model the stock returns Rt := Pt Pt 1 Pt 1 as coming from a GARCH(1,1) design. ... using least-squares regression to fit the model COs,i βt 0 + j=1 λt j COs 7j,i + j=1 κt j Fs 7j,i, s = t 14 . . . , t, i = 1, . . . , 3243.