High-dimensional multivariate forecasting with low-rank Gaussian Copula Processes

Authors: David Salinas, Michael Bohlke-Schneider, Laurent Callot, Roberto Medico, Jan Gasthaus

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

Reproducibility Variable Result LLM Response
Research Type Experimental We show on several realworld datasets that our method provides significant accuracy improvements over state-of-the-art baselines and perform an ablation study analyzing the contributions of the different components of our model.
Researcher Affiliation Collaboration David Salinas Naverlabs david.salinas@naverlabs.com Michael Bohlke-Schneider Amazon Research bohlkem@amazon.com Laurent Callot Amazon Research lcallot@amazon.com Roberto Medico Ghent University roberto.medico91@gmail.com Jan Gasthaus Amazon Research gasthaus@amazon.com
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes The code to perform the evaluations of our methods and different baselines is available at https://github.com/mbohlkeschneider/gluonts/tree/mv_release.
Open Datasets Yes The following publicly-available datasets are used to compare the accuracy of different multivariate forecasting models. Exchange rate: daily exchange rate between 8 currencies as used in [16] Solar: hourly photo-voltaic production of 137 stations in Alabama State used in [16] Electricity: hourly time series of the electricity consumption of 370 customers [6] Traffic: hourly occupancy rate, between 0 and 1, of 963 San Francisco car lanes [6] Taxi: spatio-temporal traffic time series of New York taxi rides [31] taken at 1214 locations every 30 minutes in the months of January 2015 (training set) and January 2016 (test set) Wikipedia: daily page views of 2000 Wikipedia pages used in [11]
Dataset Splits No The paper mentions splitting data into training and test sets but does not explicitly describe a separate validation split or how it was used (e.g., for hyperparameter tuning) with specific details like percentages or counts.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies Yes GARCH, a multivariate conditional heteroskedasticity model proposed by [34] with implementation from [12]. More details about these methods can be found in the supplement. [12] Alexios Ghalanos. rmgarch: Multivariate GARCH models., 2019. R package version 1.3-6.
Experiment Setup Yes In all our low-rank experiments we use r = 10. we use B = 20. we fix the context length hyperparameter T to τ. The number of past observations m used to estimate the empirical CDFs is an hyperparameter and left constant in our experiments with m = 100. For evaluation, we generate 400 samples from each model.