Fair and Optimal Decision Trees: A Dynamic Programming Approach
Authors: Jacobus van der Linden, Mathijs de Weerdt, Emir Demirović
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our results show that our model can find fair and optimal trees several orders of magnitude faster than previous methods, and now also for larger datasets that were previously beyond reach. Moreover, we show that with this substantial improvement our method can find the full Pareto front in the trade-off between accuracy and fairness. |
| Researcher Affiliation | Academia | Jacobus G.M. van der Linden Mathijs M. de Weerdt Emir Demirovi c Delft University of Technology, Department of Computer Science {J.G.M.vander Linden, M.M.de Weerdt, E.Demirovic}@tudelft.nl |
| Pseudocode | Yes | Pseudocode. This section gives the pseudocode for DPF. In addition to the description above, the pseudocode also considers upper and lower bounds and cache. Therefore, the function TF receives the current best known upper bound ub on the misclassification score as an extra parameter. ... see Algorithm 1 for the resulting pseudocode of DPF. |
| Open Source Code | No | The link to the source code will be made publicly available upon acceptance |
| Open Datasets | Yes | We evaluate DPF on all datasets mentioned in the survey by Le Quy et al. [30]. The data preprocessing and binarization is also done as described in that paper. The binarized datasets are included in our code repository (if the license permits redistribution). |
| Dataset Splits | No | The paper states: 'The algorithm receives the whole dataset as input and is asked to find a tree with an unfairness limit of δ = 1%.' It does not specify train/validation/test dataset splits, percentages, or explicit splitting methodology for the main experiments in the paper. |
| Hardware Specification | Yes | All experiments are run on a 2.6Ghz Intel i7 CPU with 8GB RAM using only one thread. |
| Software Dependencies | Yes | In our experiments, the Fair OCT model is solved with Gurobi 9.0 using the default parameters. |
| Experiment Setup | Yes | The algorithm receives the whole dataset as input and is asked to find a tree with an unfairness limit of δ = 1%. In a later experiment, we also examine the impact of this parameter on the runtime. The values shown are the average of five runs and a time limit of one hour is set for every run. All experiments are run on a 2.6Ghz Intel i7 CPU with 8GB RAM using only one thread. |