Probabilistic Forecasting: A Level-Set Approach

Authors: Hilaf Hasson, Bernie Wang, Tim Januschowski, Jan Gasthaus

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical experiments show that our approach, equipped with tree-based models as the point estimator, rivals state-of-the-art deep learning models in terms of forecasting accuracy. ... We compare LSF with the state-of-the-art models in both tabular (see Section 6.1) and forecasting tasks (See Section 6.2), and the empirical results verify the effectiveness of the proposed approach. ... In Section 6.1 we compare XLSF (LSF with XGBoost) against QRFs (as a leading tree-based probabilistic algorithm) and Conformalized Predictions (as a baseline for turning point estimators into probabilistic ones). In Section 6.2 we apply LSF to time series data. All experiments are done using Amazon Sagemaker [21] with instance type ml.m4.16xlarge.
Researcher Affiliation Industry Hilaf Hasson Amazon Research hashilaf@amazon.com Yuyang Wang Amazon Research yuyawang@amazon.com Tim Januschowski Amazon Research tjnsch@amazon.com Jan Gasthaus Amazon Research gasthaus@amazon.com
Pseudocode Yes Algorithm 1: Level Set Partitioning Algorithm ... Algorithm 2: Mean and Quantile Regression Forests with the CART-Splitting Criterion
Open Source Code Yes LSF is implemented in Gluon TS: https://github.com/awslabs/gluon-ts/blob/master/src/gluonts/model/rotbaum/README_LSF.ipynb
Open Datasets Yes Benchmarking datasets. ... electricity [13] ... parts [34] ... m4_daily [24] ... traffic [13] ... wiki10k ... dcrideshare [8]. [8] Capital Bikeshare. Capital bikeshare data, 2021. URL https://www.capitalbikeshare.com/system-data. [13] Dua Dheeru and EfiKarra Taniskidou. UCI machine learning repository, 2017. URL http://archive.ics.uci.edu/ml. [24] Spyros Makridakis, Evangelos Spiliotis, and Vassilios Assimakopoulos. The m4 competition: Results, findings, conclusion and way forward. International Journal of Forecasting, 34(4): 802 808, 2018. [34] Ralph D Snyder, J Keith Ord, and Adrian Beaumont. Forecasting the intermittent demand for slow-moving inventories: A modelling approach. International Journal of Forecasting, 28(2): 485 496, 2012.
Dataset Splits No The paper mentions training data and test data, but no explicit statement or details are provided for a validation set split or cross-validation strategy for the main experiments.
Hardware Specification Yes All experiments are done using Amazon Sagemaker [21] with instance type ml.m4.16xlarge.
Software Dependencies No The paper mentions using XGBoost [11], lightgbm ([20]), and Gluon TS (implementation link provided), but it does not specify exact version numbers for these software dependencies, which is required for reproducibility.
Experiment Setup Yes All hyperparameters used are specified in Appendix F.