Learning to Reconstruct Missing Data from Spatiotemporal Graphs with Sparse Observations

Authors: Ivan Marisca, Andrea Cini, Cesare Alippi

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate our method on three real-world datasets and compare the performance against state-of-the-art methods and standard approaches for MTSI. and Table 1 reports experimental results. Both SPIN methods outperform the baselines in almost all scenarios.
Researcher Affiliation Academia Ivan Marisca 1The Swiss AI Lab IDSIA, Università della Svizzera italiana ivan.marisca@usi.ch, Andrea Cini 1The Swiss AI Lab IDSIA, Università della Svizzera italiana andrea.cini@usi.ch, Cesare Alippi 1The Swiss AI Lab IDSIA, Università della Svizzera italiana 2Politecnico di Milano cesare.alippi@usi.ch
Pseudocode No The paper describes the model components and process in text and mathematical equations, but does not include a clearly labeled pseudocode or algorithm block.
Open Source Code Yes The code to reproduce the experiments of the paper is available online3. https://github.com/Graph-Machine-Learning-Group/spin
Open Datasets Yes We consider three openly available datasets coming from real-world SNs. The first two, namely PEMS-BAY and METR-LA [2], are two widely used benchmarks in spatiotemporal forecasting literature. ... As a third dataset, we consider AQI [59]
Dataset Splits No The paper states, 'All the baselines have been implemented in Py Torch [57] using the Torch Spatiotemporal library2 [58] and, whenever possible, open-source code provided by the authors. The code to reproduce the experiments of the paper is available online3. Please refer to the appendix for more details about the experimental setup.' but does not explicitly state training, validation, and test dataset splits with percentages or counts in the main text.
Hardware Specification No The paper states, 'The authors wish to thank the Institute of Computational Science at USI for granting access to computational resources.' and the ethics statement indicates hardware details are in supplementary material, but no specific hardware (e.g., GPU/CPU models, memory) is mentioned in the main text.
Software Dependencies No The paper mentions 'All the baselines have been implemented in Py Torch [57] using the Torch Spatiotemporal library2 [58]', but does not provide specific version numbers for these software dependencies.
Experiment Setup No The paper states 'Please refer to the appendix for more details about the experimental setup.' but does not provide specific hyperparameters (e.g., learning rate, batch size, epochs) or detailed training configurations in the main text.