Causal Dependence Plots

Authors: Joshua Loftus, Lucius Bynum, Sakina Hansen

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate with simulations and real data experiments how CDPs can be combined in a modular way with methods for causal learning or sensitivity analysis.
Researcher Affiliation Academia Joshua R. Loftus Department of Statistics London School of Economics London, England, UK j.r.loftus@lse.ac.uk Lucius E. J. Bynum Center for Data Science New York University New York, NY, USA lucius@nyu.edu Sakina Hansen Department of Statistics London School of Economics London, England, UK s.a.hansen1@lse.ac.uk
Pseudocode Yes CDP pseudo-algorithm. To construct a CDP showing how ˆf(x) depends on xj, a user specifies an ECM M containing the predictors x, and an intervention I(xj) in M. ... Algorithm 1 Total Dependence Plot (TDP) ... Algorithm 2 Explanatory Causal Model (ECM) ... Algorithm 3 Partially Controlled Dependence Plot (PCDP) ... Algorithm 4 Natural Direct Dependence Plot (NDDP) ... Algorithm 5 Natural Indirect Dependence Plot (NIDP)
Open Source Code Yes Our code to implement CDPs, run the experiments, and produce figures is available at this repository: https://github.com/causalhypothesis/cdp-neurips/
Open Datasets Yes The Breast Cancer Wisconsin (Original) dataset [35] is a publicly available dataset often used to test algorithms on medical data. ... Sachs et al. [49] dataset.
Dataset Splits No The paper mentions 'explanatory dataset' and 'training data' but does not specify explicit training/validation/test splits (e.g., percentages or sample counts) for its experiments. For example, it states: 'Points show the explanatory dataset, which in this example is also the training data for the predictive models.'
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. It only states: 'Our experiments are not computationally costly and can be reproduced on a personal computer.'
Software Dependencies No The paper mentions several software packages used: 'Predictive models were fit using scikit-learn [42]. CDP implementations make use of causal modeling functions in dowhy [53]. Figures were generated with matplotlib [20]. For causal structural learning we used the PC algorithm [55] implemented in Julia Causal Inference [51]. We used the Python implementation of ALE plots in [22].' However, it does not specify version numbers for these software components.
Experiment Setup No The paper describes the general setup of experiments (e.g., types of models used, datasets) and mentions that 'full details are given in supplemental material' for reproducibility. However, it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs) or detailed system-level training settings within the main text of the paper.