Discovering Sequential Patterns with Predictable Inter-event Delays

Authors: Joscha Cüppers, Paul Krieger, Jilles Vreeken

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that HOPPER efficiently recovers the ground truth, discovers meaningful patterns from real-world data, and outperforms existing methods in discovering long-delay patterns. In this section we empirically evaluate HOPPER on synthetic and real-world data.
Researcher Affiliation Academia Joscha C uppers1, Paul Krieger2, Jilles Vreeken1 1 CISPA Helmholtz Center for Information Security 2 Saarland University
Pseudocode Yes We give the pseudocode as Algorithm 1. Algorithm 2: HOPPER
Open Source Code Yes We make all code, synthetic data, and real-world datasets available in the supplementary material.1 [1eda.rg.cispa.io/prj/hopper]
Open Datasets Yes We implement HOPPER in Python and provide the source code along with the synthetic data and the real-world data in the supplementary.1 We use eight datasets that together span a wide range of use-cases. We consider a dataset of all national Holidays in a European country over a century, the playlist a local Radio station recorded over a month, the Lifelog2 of all activities of one person recorded in over seven years, the MIDI data of hundred Bach Chorales (Dua and Graff 2017), all commits to the Samba project for over ten years (Galbrun et al. 2018), the Rolling Mill production log of steel manufacturing plant (Wiegand, Klakow, and Vreeken 2021), the discretized muscle activations of professional ice Skating riders (Moerchen and Fradkin 2010), and finally, three text datasets the Gutenberg project, resp. Romeo and Juliet by Shakespeare, A Room with a View by E.M. Forster, and The Great Gatsby by F. Scott Fitzgerald.
Dataset Splits No No specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology for training, validation, and testing) is provided in the paper.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory specifications) used for running the experiments are mentioned in the paper.
Software Dependencies No The paper states "We implement HOPPER in Python" but does not provide specific version numbers for Python or any other software libraries or solvers used.
Experiment Setup Yes HOPPER considers delays up to a user set max delay, for all experiments we set it to 200. SKOPUS only works on a set of sequences, when the dataset consists of one sequence, we split the sequence into 100 equally long sequences. We parametrize SKOPUS to report the top 10 patterns of at most length 10, which corresponds to the ground-truth value in our synthetic experiments. PPM only accepts a single sequence as input, to make it applicable on databases of multiple sequences, we concatenate these into one long sequence. We give the full setup description in the supplementary.