Fully-Dynamic Decision Trees

Authors: Marco Bressan, Gabriel Damay, Mauro Sozio

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We complement our theoretical results with an extensive experimental evaluation on real-world data, showing the effectiveness of our algorithm. Our contributions can be summarized as follows: We conduct an extensive experimental evaluation on real-world data, evaluating FUDYADT s speed and accuracy against state-of-the-art tools such as EFDT and HAT.
Researcher Affiliation Academia 1University of Milan, 2Institut Polytechnique de Paris, T el ecom Paris marco.bressan@unimi.it, gabriel.damay@telecom-paris.fr, sozio@telecom-paris.fr
Pseudocode Yes Algorithm 1 FUDYADT.UPDATE; Algorithm 2 FUDYADT.BUILD
Open Source Code Yes We implemented FUDYADT in C++ 2. (Footnote 2: https://github.com/GDamay/dynamic-tree)
Open Datasets Yes Our datasets are shown in Table 1. (Table 1 lists: Electricity, Forest Covertype, INSECTS v1-v5, KDDCUP99, NOAA Weather, Poker)
Dataset Splits No The paper describes a sequential processing approach and mentions a 'grace period' for baseline parameter tuning, but it does not specify explicit train/validation/test splits with percentages or counts for its own experiments. The concept of a distinct 'validation' set is not explicitly defined for model training or selection.
Hardware Specification Yes We conducted all experiments on an Ubuntu 20.04.2 LTS server equipped with 144 Intel(R) Xeon(R) Gold 6154 @ 3.00GHz CPUs and 264 GB of RAM.
Software Dependencies No The paper states: 'We implemented FUDYADT in C++' and 'using the MOA software'. However, it does not provide specific version numbers for C++ compiler, MOA software, or any other libraries used, which are required for full reproducibility.
Experiment Setup Yes For FUDYADT, we let α = 0, β = 0 k = 1, h ∈ {5, 10}, and we manually set ϵ ∈ [0, 2]. For the SW model we use W ∈ {100, 1000}. The parameters of EFDT and HAT are set to the original values specified by the authors; we only vary the so-called grace period in {100, 500, 1000} to find the value yielding highest F1-score.