Adaptive Gaussian Process Change Point Detection

Authors: Edoardo Caldarelli, Philippe Wenk, Stefan Bauer, Andreas Krause

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

Reproducibility Variable Result LLM Response
Research Type Experimental In extensive experiments, we show its versatility and applicability to different classes of change points, demonstrating that it is significantly more accurate than current state-of-the-art alternatives. This section reports the empirical results obtained using ADAGA in the context of CP detection in time series.
Researcher Affiliation Academia 1Institut de Rob otica i Inform atica Industrial, CSIC-UPC, Barcelona, Spain 2Department of Computer Science, ETH Zurich, Zurich, Switzerland 3Department of Intelligent Systems, KTH, Stockholm, Sweden.
Pseudocode Yes Algorithm 1 ADAGA
Open Source Code Yes Code available at https://github.com/lasgroup/adaga. The implementation uses scipy (Virtanen et al., 2020), tensorflow (Abadi et al., 2016) and gpflow (De G. Matthews et al., 2017).
Open Datasets Yes Secondly, we use some of the real-world benchmarks described by van den Burg & Williams (2020). In our work, we choose the following sets, whose detailed description is reported in the supplementary material: Run Log, Business Inventories, Ozone, Iran s GDP, Argentina s GDP, Japan s GDP.
Dataset Splits No The paper does not explicitly specify the training/validation/test dataset splits (e.g., percentages, sample counts, or references to predefined splits) for its own experimental setup.
Hardware Specification Yes All the experiments were performed on a custom laptop (Macbook Pro 2019).
Software Dependencies No The implementation uses scipy (Virtanen et al., 2020), tensorflow (Abadi et al., 2016) and gpflow (De G. Matthews et al., 2017). The paper provides software names but not explicit version numbers for reproducibility, only citations to their original papers.
Experiment Setup Yes The minimum window size |S| for ADAGA is set to 15 points. We run ADAGA with a batch size of 1, to achieve maximum resolution in the location of the CPs. To avoid overly conservative estimates, we set δ = 0.6.