Antonio De Leon Bayesian Statistics | Machine Learning | Statistical Software

Six-part series

Bayesian workflow

A compact sequence from problem formulation through computation, simulation, model criticism, predictive comparison, and final reporting.

  1. 1 From questions to a generative model Sep 29, 2024 Updated Jul 29, 2026 4 min read Bayesian analysis starts with a target, a data-generating story, and assumptions that can be inspected before fitting.
  2. 2 Preparing and monitoring computation Oct 11, 2024 Updated Jul 29, 2026 4 min read Computation should be staged, diagnosed, and recorded against the quantities that will actually be reported.
  3. 3 Simulation, recovery, and calibration Nov 1, 2024 Updated Jul 29, 2026 4 min read Simulation helps separate coding errors, weak identification, poor priors, and biased computation before real-data conclusions are trusted.
  4. 4 Diagnosing fit and revising the model Nov 13, 2024 Updated Jul 29, 2026 4 min read A failed diagnostic should lead to a specific revision: code, computation, parameterization, prior information, or model structure.
  5. 5 Comparing and combining predictive models Dec 2, 2024 Updated Jul 29, 2026 3 min read Model comparison should match the prediction task, report uncertainty, and avoid treating stacking weights as posterior model probabilities.
  6. 6 Iteration, stopping, and reporting Dec 19, 2024 Updated Jul 29, 2026 4 min read A model is ready to report when it answers the target with acceptable uncertainty, not when every possible discrepancy has disappeared.

Methods and computing

Standalone notes

Scientific computing 5 min read

A practical Slurm workflow on UCSC Hummingbird

Use Hummingbird by discovering current partitions, testing small interactive jobs, controlling threads, and submitting reproducible Slurm batches.

Updated