N-BEATS: Neural basis expansion analysis for interpretable time series forecasting
Authors: Boris N. Oreshkin, Dmitri Carpov, Nicolas Chapados, Yoshua Bengio
ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We test the proposed architecture on several well-known datasets, including M3, M4 and TOURISM competition datasets containing time series from diverse domains. We demonstrate state-of-the-art performance for two configurations of N-BEATS for all the datasets, improving forecast accuracy by 11% over a statistical benchmark and by 3% over last year s winner of the M4 competition... and 5 EXPERIMENTAL RESULTS Our key empirical results based on aggregate performance metrics over several datasets M4 (M4 Team, 2018b; Makridakis et al., 2018b), M3 (Makridakis & Hibon, 2000; Makridakis et al., 2018a) and TOURISM (Athanasopoulos et al., 2011) appear in Table 1. |
| Researcher Affiliation | Collaboration | Boris N. Oreshkin Element AI boris.oreshkin@gmail.com Dmitri Carpov Element AI dmitri.carpov@elementai.com Nicolas Chapados Element AI chapados@elementai.com Yoshua Bengio Mila yoshua.bengio@mila.quebec |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. It describes the architecture and operations using text and mathematical equations. |
| Open Source Code | No | The paper states 'N-BEATS is implemented and trained in Tensorflow (Abadi et al., 2015).' and cites the TensorFlow URL, but does not provide any concrete access to the specific source code for the N-BEATS methodology described in the paper. |
| Open Datasets | Yes | M4 (M4 Team, 2018b; Makridakis et al., 2018b) is the latest in an influential series of forecasting competitions organized by Spyros Makridakis since 1982 (Makridakis et al., 1982). The 100k-series dataset is large and diverse... M4 Team. M4 dataset, 2018a. URL https://github.com/M4Competition/M4-methods/tree/master/Dataset. M3 (Makridakis & Hibon, 2000)... TOURISM (Athanasopoulos et al., 2011) dataset was released as part of the respective Kaggle competition conducted by Athanasopoulos & Hyndman (2011). |
| Dataset Splits | Yes | We split each dataset into train, validation and test subsets. The test subset is the standard test set previously defined for each dataset (M4 Team, 2018a; Makridakis & Hibon, 2000; Athanasopoulos et al., 2011). The validation and train subsets for each dataset are obtained by splitting their full train sets at the boundary of the last horizon of each time series. We use the train and validation subsets to tune hyperparameters. |
| Hardware Specification | No | The paper mentions 'The GPU based training of one ensemble member for entire M4 dataset takes between 30 min and 2 hours depending on neural network settings and hardware,' but does not provide specific details such as GPU models, CPU types, or other hardware specifications. |
| Software Dependencies | No | The paper states 'N-BEATS is implemented and trained in Tensorflow (Abadi et al., 2015).', but it does not specify the version number for TensorFlow or any other software dependencies. |
| Experiment Setup | Yes | Table 18: Settings of hyperparameters across subsets of M4, M3, TOURISM datasets. ... Parameter N-BEATS-I: LH, Iterations, Losses, S-width, S-blocks, S-block-layers, T-width, T-degree, T-blocks, T-block-layers, Sharing, Lookback period, Batch. Parameter N-BEATS-G: LH, Iterations, Losses, Width, Blocks, Block-layers, Sharing, Lookback period, Batch. |