[2402.16326] A Provably Correct Randomized Sampling Algorithm for Logistic Regression


Obtain a PDF of the paper titled A Provably Correct Randomized Sampling Algorithm for Logistic Regression, by Agniva Chowdhury and 1 different authors

Obtain PDF
HTML (experimental)

Summary:In statistics and machine studying, logistic regression is a widely-used supervised studying method primarily employed for binary classification duties. When the variety of observations enormously exceeds the variety of predictor variables, we current a easy, randomized sampling-based algorithm for logistic regression drawback that ensures high-quality approximations to each the estimated chances and the general discrepancy of the mannequin. Our evaluation builds upon two easy structural circumstances that boil right down to randomized matrix multiplication, a elementary and well-understood primitive of randomized numerical linear algebra. We analyze the properties of estimated chances of logistic regression when leverage scores are used to pattern observations, and show that correct approximations could be achieved with a pattern whose dimension is far smaller than the full variety of observations. To additional validate our theoretical findings, we conduct complete empirical evaluations. General, our work sheds mild on the potential of utilizing randomized sampling approaches to effectively approximate the estimated chances in logistic regression, providing a sensible and computationally environment friendly answer for large-scale datasets.

Submission historical past

From: Agniva Chowdhury [view email]
Mon, 26 Feb 2024 06:20:28 UTC (10,841 KB)
Thu, 29 Feb 2024 09:05:38 UTC (10,841 KB)

Supply hyperlink


Please enter your comment!
Please enter your name here