Spectral Subsampling MCMC for Stationary Time Series

Authors: Robert Salomone, Matias Quiroz, Robert Kohn, Mattias Villani, Minh-Ngoc Tran

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We consider several time series models for large data in our experiments, including the recently proposed class of autoregressive tempered fractionally integrated moving average (ARTFIMA) models and several of its widely-used special cases such as ARMA and ARFIMA though we highlight that any model class for which the spectral density is known can be used.
Researcher Affiliation Academia 1UNSW Sydney 2University of Technology Sydney 3Stockholm University 4Link oping University 5University of Sydney.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about making the source code available or links to a code repository for the methodology described.
Open Datasets Yes Example 1: Vancouver Temperatures (ARMA) sourced from openweathermap.org. Example 2: Stockholm Temperatures (ARTFIMA) obtained from the Swedish Meteorological and Hydrological Institute (www.smhi.se/en). Example 4: Bitcoin Prices Stochastic Volatility (ARTFIMA-SV) from coinbase.com.
Dataset Splits No The paper mentions dataset lengths (e.g., 'length n = 44001') but does not specify training, validation, or test splits by percentage or sample count. No explicit split methodology is provided.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments (e.g., GPU/CPU models, memory).
Software Dependencies No The paper mentions 'CODA package (Plummer et al., 2006) in R' and 'auto.arima function in the Forecast (Hyndman & Khandakar, 2008) package in R', but does not provide specific version numbers for R, CODA, or Forecast, which are required for reproducibility.
Experiment Setup Yes Table 1. Settings for models ARMA(2,3) (M1), ARTFIMA(2,2) (M2), ARFIMA(2,1) (M3) and ARTFIMA-SV(1,1) (M4). The table shows number of frequency observations (n), number of groups (|G|), number of observations per group (|G|, the same for all groups), percentage of subsampled groups (m) and number of blocks in the block pseudo-marginal algorithm (B). The coreset settings are number of iterations of GIGA algorithm (M), number of random projections (RP) (see Campbell & Broderick (2018) for details) and the average size of the coreset (gk).