Structural Maxent Models

Authors: Corinna Cortes, Vitaly Kuznetsov, Mehryar Mohri, Umar Syed

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental report the results of several experiments demonstrating that its performance improves on that of existing L1-norm regularized Maxent algorithms. In Section 4, we report the results of extensive experiments with data from various domains including community crimes, traffic and species habitat modeling.
Researcher Affiliation Collaboration Corinna Cortes CORINNA@GOOGLE.COM Google Research, 111 8th Avenue, New York, NY 10011 Vitaly Kuznetsov VITALY@CIMS.NYU.EDU Courant Institute of Mathematical Sciences, 251 Mercer Street, New York, NY 10012 Mehryar Mohri MOHRI@CIMS.NYU.EDU Courant Institute and Google Research, 251 Mercer Street, New York, NY 10012 Umar Syed USYED@GOOGLE.COM Google Research, 111 8th Avenue, New York, NY 10011
Pseudocode Yes Figure 1. Pseudocode of the Struct Maxent2 algorithm.
Open Source Code No We have fully implemented both Struct Maxent1 and Struct Maxent2 with diverse feature families and will make the software used in our experiments available as open-source.
Open Datasets Yes For our experiments, we used two data sets from (Phillips et al., 2006) which are accessible from http://www.cs.princeton.edu/ schapire/maxent. We also experimented with a Minnesota traffic dataset (Kwon, 2004) which is accessible from http://www.d.umn.edu/ tkwon/TMCdata/TMCarchive.html. We used the UCI Communities and Crime dataset as another test case for Struct Maxent algorithm.
Dataset Splits Yes For each species, we first randomly split the sample S into a training set S1 (70%) and a test set S2 (30%). We trained all algorithms on S1 and used the error on S2 to find the optimal value of the parameters λ and β.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper states that software was implemented but does not provide specific ancillary software details (e.g., library or solver names with version numbers).
Experiment Setup Yes Specifically, we optimized over λ, β {0.0001, 0.001, 0.01, 0.1, 0.5, 1, 2}. We compared Struct Maxent with L1-regularized Maxent...we ran each algorithm for 500 rounds, or until the change in the objective on a single round fell below 10^-5.