Branches: Efficiently Seeking Optimal Sparse Decision Trees via AO*

Authors: Ayman Chaouki, Jesse Read, Albert Bifet

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We prove both optimality and complexity guarantees for BRANCHES and we show that it is more efficient than the state of the art theoretically and on a variety of experiments. (Abstract). We employ 11 datasets from the UCI repository. For each dataset we use different types of encodings... (Section 5, Experiments)
Researcher Affiliation Academia 1LIX, Ecole Polytechnique, IP Paris 2AI Institute, University of Waikato 3LTCI, T el ecom Paris, IP Paris.
Pseudocode Yes We provide implementation details and a pseudocode in Appendix D and Algorithm 1.
Open Source Code Yes BRANCHES https://github.com/Chaoukia/branches.
Open Datasets Yes We employ 11 datasets from the UCI repository.
Dataset Splits Yes Figure 13, Figure 14, Figure 15 and Figure 16 report the median and the quartiles of the accuracy and number of splits of the proposed solutions induced by a 10 fold crossvalidation
Hardware Specification Yes we run all the experiments on a personal Machine (Processor: 2,6 GHz 6-Core Intel Core i7; Memory: 16 GB).
Software Dependencies No No specific software dependencies with version numbers are explicitly listed for the experiments. The paper mentions a "Python implementation" for BRANCHES, but does not provide version numbers for Python or any specific libraries used.
Experiment Setup Yes We set a time limit of 5 minutes for all algorithms and we run all the experiments on a personal Machine (Processor: 2,6 GHz 6-Core Intel Core i7; Memory: 16 GB). Moreover, since it is necessary to fix a maximum depth for DFS methods, we set it to 20 for Mur Tree and STree D while the BFS methods GOSDT and BRANCHES run with infinite maximum depth. The λ values reported in Table 4 are those employed in the experiments of Section 5, they were chosen from a pool of values λ {0.1, 0.05, 0.025, 0.01, 0.005, 0.001} to yield meaningful solutions. For DL8.5, we ran it with with maximum depths raging between 1 and 10. We set the maximum nodes for Mur Tree to 80.