Discovering Context Effects from Raw Choice Data

Authors: Arjun Seshadri, Alex Peysakhovich, Johan Ugander

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4. Experiments We now evaluate the CDM and low-rank CDM on data. Our evaluation includes comparisons with MNL/Luce models and mixed MNL models (Mc Fadden and Train, 2000). MNL and CDM model likelihoods are optimized using Adam (Kingma and Ba, 2014), a stochastic gradient descent algorithm with adaptive moment estimation. Mixed MNL likelihoods are optimized using open source code from (Ragain and Ugander, 2016). The CDM parameter optimization is initialized with values corresponding to a Luce MLE for that dataset.
Researcher Affiliation Collaboration 1Stanford University, Stanford, CA 2Facebook Artificial Intelligence Research, New York, NY.
Pseudocode No The paper describes the Context Dependent Random Utility Model (CDM) mathematically and textually, but does not include any explicit pseudocode blocks or algorithm listings.
Open Source Code No Replication code for all figures will be released at publication time.
Open Datasets Yes We now turn to two real-world datasets: SFwork and SFshop. These data are collected from a survey of transportation preferences around the San Francisco Bay Area (Koppelman and Bhat, 2006).
Dataset Splits No The paper mentions evaluating 'out of sample fit on a held out 20% of the data' for the SFwork/SFshop datasets and 'on a 20% held-out test set' for the nature photo dataset. It describes a train/test split but does not explicitly provide details for a separate validation set for hyperparameter tuning or model selection.
Hardware Specification No The paper mentions the use of Adam for optimization and other open-source code but does not provide specific details on the hardware (e.g., GPU/CPU models, memory specifications) used for running the experiments.
Software Dependencies No The paper mentions 'Adam' as an optimizer and 'open source code from (Ragain and Ugander, 2016)' but does not provide specific version numbers for any software dependencies or libraries used in the experiments.
Experiment Setup Yes The CDM parameter optimization is initialized with values corresponding to a Luce MLE for that dataset.