MERCS: Multi-Directional Ensembles of Regression and Classification Trees

Authors: Elia Van Wolputte, Evgeniya Korneva, Hendrik Blockeel

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show the viability of the approach. This section is divided in three parts. First, we study the selection strategies. Second, we investigate the effects of the different prediction strategies and third, we compare MERCS to an external baseline, in this case PGMs.
Researcher Affiliation Academia Elia Van Wolputte, Evgeniya Korneva, Hendrik Blockeel KU Leuven, Department of Computer Science, 3001 Leuven, Belgium {elia.vanwolputte, evgeniya.korneva, hendrik.blockeel}@cs.kuleuven.be
Pseudocode Yes Algorithm 1 Prediction algorithm
Open Source Code No The paper does not explicitly state that the source code for MERCS is publicly available nor does it provide a link to a code repository. It mentions using 'the default Sci Kit-Learn implementation of decision trees'.
Open Datasets Yes The datasets used in the following experiments are summarized in Table 2. All datasets previously occurred in studies on density estimation (also inherently a multidirectional setting), except for nursery and adult, which were lifted from the UCI repository.
Dataset Splits No The paper does not explicitly provide specific training, validation, and testing dataset splits (e.g., percentages, sample counts, or citations to predefined splits). It describes generating 'queries' for evaluation, but this is distinct from traditional data splits.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU/GPU models, memory, or cloud instance types).
Software Dependencies No The paper mentions using the 'SMILE library' and 'the default Sci Kit-Learn implementation of decision trees', but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes The selection strategy used in this paper is the following: construct q random partitions of A; for each partition Pi = {Yi,1, Yi,2, . . . , Yi, m p }, learn for each Yi,j k trees T(A \ Yi,j, Yi,j). Thus, the total number of trees included in the model is m p kq. The threshold T (Alg. 1) that Capp has to exceed to determine appropriateness can vary. For Bayes Fusion this means greedy thick thinning for structure learning and the clustering algorithm for inference. For MERCS, we rely on the default Sci Kit-Learn implementation of decision trees.