Effective Bayesian Modeling of Groups of Related Count Time Series

Authors: Nicolas Chapados

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate model performance on three datasets obtained from supply chain operations... Model performance is evaluated by the out-of-sample forecasting accuracy over the horizon h (where h varies from 1 to 12 periods), measured according to the negative loglikelihood (NLL) per period, relative mean squared error (MSE) and relative mean absolute error (MAE).
Researcher Affiliation Industry Nicolas Chapados CHAPADOS@APSTAT.COM Ap STAT Technologies Inc., 408-4200 Boul. St-Laurent, Montral, QC, H2W 2R2, CANADA
Pseudocode No The paper describes algorithmic steps in narrative form but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper mentions using and implementing some benchmarks (e.g., R forecast package, Stan modeling language) but does not state that the code for the authors' H-NBSS model is open-source or publicly available.
Open Datasets No The paper identifies datasets used (RAID, GLUE, PARTS) and states they were 'obtained from supply chain operations' or 'previously studied by Syntetos et al. (2012)', but it does not provide concrete access information (e.g., direct links, DOIs, or full citations with author and year in brackets/parentheses for public repositories) for these datasets.
Dataset Splits Yes Model performance is evaluated by a sequential re-training procedure that alternates between model training and testing, moving at each iteration the first observation of the (previous) test set to the end of the (new) training set... The initial training set durations for each dataset are given in Table 1.
Hardware Specification No The paper discusses computational time and efficiency but does not specify the hardware (e.g., CPU, GPU models, memory) used for the experiments.
Software Dependencies Yes Our implementation of the Laplace approximation (coded in the interpreted language R)... equivalent model computed with Markov chain Monte Carlo (MCMC)... implemented in the Stan modeling language (Stan Development Team, 2013)... implemented by the corresponding functions in the R forecast package (Hyndman et al., 2013).
Experiment Setup Yes a reasonable initialization for ηℓ,t, 1 t T h can be taken as the midpoint between between log yℓ,t and the mean of log-values for series ℓ, mℓ. Initial values for T h + 1 t T can be taken to be mℓ... We combined the results of four independent chains, each run with 1500 burn-in iterations followed by 18500 sampling iterations.