Precision and Recall for Time Series

Authors: Nesime Tatbul, Tae Jun Lee, Stan Zdonik, Mejbah Alam, Justin Gottschlich

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present an experimental study of our range-based model applied to results of state-of-the-art AD algorithms over a collection of diverse time series datasets.
Researcher Affiliation Collaboration Nesime Tatbul Intel Labs and MIT tatbul@csail.mit.edu Tae Jun Lee Microsoft tae_jun_lee@alumni.brown.edu Stan Zdonik Brown University sbz@cs.brown.edu Mejbah Alam Intel Labs mejbah.alam@intel.com Justin Gottschlich Intel Labs justin.gottschlich@intel.com
Pseudocode Yes Figure 2: Example definitions for ω() and δ() functions. (a) Overlap size function ω(Anomaly Range, Overlap Set, δ) (b) Positional bias function δ(i, Anomaly Length)
Open Source Code No The paper provides links to third-party tools (NAB Data Corpus [32], Paranom tool [15], Luminol [28]) used in their experiments, but does not state that the source code for their own proposed model or implementations is open-source or available.
Open Datasets Yes The real datasets are taken from the NAB Data Corpus [32], whereas the synthetic datasets are generated by the Paranom tool [15]. ... [32] Numenta. NAB Data Corpus. https://github.com/numenta/NAB/tree/master/data/, 2017. ... [15] J. Gottschlich. Paranom: A Parallel Anomaly Dataset Generator. https://arxiv.org/abs/1801. 03164/, 2018.
Dataset Splits No The paper states: "For training and testing, we carefully partition each dataset..." but does not specify details about a validation split or its size/methodology.
Hardware Specification Yes System: All experiments were run on a Windows 10 machine with an Intel R Core TM i5-6300HQ processor running at 2.30 GHz with 4 cores and 8 GB of RAM.
Software Dependencies No The paper mentions software like "TensorFlow-implemented version of LSTM-AD", "Greenhouse [26]", and "Luminol [28]" but does not provide specific version numbers for any of these software dependencies.
Experiment Setup Yes Unless otherwise stated, we use the following default parameter settings for computing our Recall T and Precision T equations: α = 0, γ() = 1, ω() is as in Figure 2a, and δ() returns flat bias as in Figure 2b.