Corpus ID: 59336361

Testing Conditional Predictive Independence in Supervised Learning Algorithms

  title={Testing Conditional Predictive Independence in Supervised Learning Algorithms},
  author={David S. Watson and Marvin N. Wright},
  • David S. Watson, Marvin N. Wright
  • Published 2019
  • Computer Science, Mathematics
  • ArXiv
  • We propose a general test of conditional independence. The conditional predictive impact (CPI) is a provably consistent and unbiased estimator of one or several features’ association with a given outcome, conditional on a (potentially empty) reduced feature set. The measure can be calculated using any supervised learning algorithm and loss function. It relies on no parametric assumptions and applies equally well to continuous and categorical predictors and outcomes. The CPI can be efficiently… CONTINUE READING
    Conditional independence testing: a predictive perspective
    • 2
    • PDF
    Floodgate: inference for model-free variable importance


    Publications referenced by this paper.
    Statistical learning theory
    • 24,877
    • PDF
    Probabilistic Graphical Models - Principles and Techniques
    • 5,129
    • PDF
    Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles
    • 20,570
    • Highly Influential
    • PDF
    Bayesian t tests for accepting and rejecting the null hypothesis
    • 2,216
    • PDF
    A Unified Approach to Interpreting Model Predictions
    • 1,375
    • PDF
    Predicting good probabilities with supervised learning
    • 682
    • PDF
    Kernel-based Conditional Independence Test and Application in Causal Discovery
    • 278
    • PDF
    A theory of the learnable
    • 3,348
    • Highly Influential
    • PDF
    On Calibration of Modern Neural Networks
    • 866
    • Highly Influential
    • PDF