$\clubsuit$ CLOVER $\clubsuit$: Probabilistic Forecasting with Coherent Learning Objective Reparameterization
Abstract: Obtaining accurate probabilistic forecasts is an operational challenge in many applications, such as energy management, climate forecasting, supply chain planning, and resource allocation.
Many of these applications present a natural hierarchical structure over the forecasted quantities; and forecasting systems that adhere to this hierarchical structure are said to be coherent.
Furthermore, operational planning benefits from the accuracy at all levels of the aggregation hierarchy. However, building accurate and coherent forecasting systems is challenging: classic multivariate time series tools and neural network methods are still being adapted for this purpose. In this paper, we augment an MQForecaster neural network architecture with a modified multivariate Gaussian factor model that achieves coherence by construction. The factor model samples can be differentiated with respect to the model parameters, allowing optimization on arbitrary differentiable learning objectives that align with the forecasting system's goals, including quantile loss and the scaled Continuous Ranked Probability Score (CRPS). We call our method the Coherent Learning Objective Reparametrization Neural Network (CLOVER). In comparison to state-of-the-art coherent forecasting methods,
CLOVER achieves significant improvements in scaled CRPS forecast accuracy, with average gains of 15%, as measured on six publicly-available datasets.
Submission Length: Regular submission (no more than 12 pages of main content)
Previous TMLR Submission Url: https://openreview.net/forum?id=B2LY1P3vrx
Changes Since Last Submission: Dear Editor and Reviewers,
We are pleased that the Action Editor and the Reviewers appreciate our manuscript. We have addressed all points raised in the reviews. We hope that you find the current version ready to be published.
In response to the feedback we have taken the opportunity to improve the paper’s main experiments, and flow. The key changes in the updated manuscript are:
1. We improved the contributions, and introduction, reducing it from the previous four pages to one. We added Section 2, dedicated to the hierarchical forecast notation.
2. We doubled the number of datasets in the main experiments in Section 4. Now the main experiments include Australian Labour, monthly Tourism, quarterly Tourism, Favorita sales, SF Traffic and Wikipedia article visits datasets.
3. We extended the ablation studies, to compare our model to well-established neural architecture alternatives, including TFT, NBEATS, NHITS, LSTM, and FC-GAGA. We extended ablation study and main experiments to include results from our model trained with the energy score.
4. We renamed our method, from DeepCoFactor to CLOVER, to better reflect the coherent learning reparameterization trick used in the paper. We fixed the final details as recommended by the Action Editor.
Sincerely yours,
The Authors
Video: https://drive.google.com/file/d/1-xkmuYSB7YQDXOEaBKeFa-pAa1Gq1-i8/view?pli=1
Code: https://github.com/Nixtla/hierarchicalforecast/tree/main/experiments/hierarchical_baselines
Assigned Action Editor: ~Mark_Coates1
Submission Number: 3148
Loading