Discovering Latent Covariance Structures for Multiple Time Series

Authors: Anh Tong, Jaesik Choi

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments conducted on five real-world data sets demonstrate that our new model outperforms existing methods in term of structure discoveries and predictive performances.
Researcher Affiliation Academia 1Department of Computer Science and Engineering, Ulsan National Institute of Science and Technology, Ulsan, 44919, South Korea. Correspondence to: Jaesik Choi <jaesik@unist.ac.kr>.
Pseudocode Yes Algorithm 1 Partial set expansion of LKM learning
Open Source Code No The paper does not provide a direct link to a code repository or an explicit statement about the public availability of its source code.
Open Datasets Yes The US stock price data set consists of 9 stocks (GE, MSFT, XOM, PFE, C, WMT, INTC, BP, and AIG) containing 129 adjusted closes taken from the second half of 2001. The US housing market data set includes the 120-month housing prices of 6 cities (New York, Los Angeles, Chicago, Phoenix, San Diego, San Francisco) from 2004 to 2013. The currency data set includes 4 currency exchange rates from US dollar to 4 emerging markets: South African Rand (ZAR), Indonesian Rupiah (IDR), Malaysian Ringgit (MYR), and Russian Rouble (RUB). Each currency exchange time series has 132 data points. We collected time series from various domains into a data set. It consists of gold prices, crude oil prices, NASDAQ composite index, and USD index1 from 2015 July 1st to 2018 July 1st. We call this data set as GONU (Gold, Oil, NASDAQ, USD index). Each time series has 157 weekly prices or indexes taken from Quandl (2018). We retrieved the epileptic seizure data set (Andrzejak et al., 2002) from UCI repository (Dheeru & Karra Taniskidou, 2017).
Dataset Splits Yes All experiments are conducted to predict future events (extrapolation) by splitting all data sets and trained with the first 90%, then tested with the remaining 10% as in the standard setting for extrapolation tasks.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper mentions general tools like GPy and Stan language in related work sections but does not list specific software dependencies with version numbers for its own implementation.
Experiment Setup Yes All experiments are conducted to predict future events (extrapolation) by splitting all data sets and trained with the first 90%, then tested with the remaining 10% as in the standard setting for extrapolation tasks. Root mean square error (RMSE) and Mean Negative Log Likelihood (MNLP) (L azaro-Gredilla et al., 2010) are the main evaluation metrics in all data sets. RMSEs and NMLPs for each data set with corresponding methods (5 independent runs per method).