Issues

• The C++ E-step and synthetic data generator count sequences and answers in int

  CAHLR/pyBKT#155 · Filter by author ChristianMurphy opened on Nov 3, 2026

• Property tests for changes to Model's input that shouldn't change its output

  CAHLR/pyBKT#150 · Filter by author ChristianMurphy opened on Nov 2, 2026

• Run the tests under AddressSanitizer and UndefinedBehaviorSanitizer in CI

  CAHLR/pyBKT#149 · Filter by author ChristianMurphy opened on Nov 2, 2026

• The C++ E-step crashes or accepts bad input when called directly

  CAHLR/pyBKT#148 · Filter by author ChristianMurphy opened on Nov 2, 2026

• synthetic_data reads outside its transition matrix when a resource ID is out of range

  CAHLR/pyBKT#147 · Filter by author ChristianMurphy opened on Nov 2, 2026

• Float32 parameters make the C++ E-step and prediction read past their arrays

  CAHLR/pyBKT#146 · Filter by author ChristianMurphy opened on Nov 2, 2026

• predict runs a backward pass it never uses

  CAHLR/pyBKT#143 · Filter by author ChristianMurphy opened on Nov 2, 2026

• The E-step keeps the whole backward pass when it needs one step at a time

  CAHLR/pyBKT#142 · Filter by author ChristianMurphy opened on Nov 2, 2026

• The C++ E-step and prediction ask for heap memory once per answer

  CAHLR/pyBKT#141 · Filter by author ChristianMurphy opened on Nov 2, 2026

• Tracking: open issues and PRs, with a suggested merge order

  CAHLR/pyBKT#140 · Filter by author ChristianMurphy opened on Nov 2, 2026

• A failed fit empties the model, duplicate index labels break predict, and C++ prediction ignores parallel=False

  CAHLR/pyBKT#132 · Filter by author ChristianMurphy opened on Nov 2, 2026

• evaluate and crossvalidate give custom metrics the missing answers and the multiprior placeholder rows

  CAHLR/pyBKT#131 · Filter by author ChristianMurphy opened on Nov 2, 2026

• fit fails when multiprior and multigs are used together

  CAHLR/pyBKT#130 · Filter by author ChristianMurphy opened on Nov 2, 2026

• With multigs, a missing answer is predicted with the first guess/slip class

  CAHLR/pyBKT#129 · Filter by author ChristianMurphy opened on Nov 2, 2026

• A missing skill name becomes a skill whose name depends on the pandas version and dtypes

  CAHLR/pyBKT#128 · Filter by author ChristianMurphy opened on Nov 2, 2026

• A second fit or crossvalidate call drops the seed, num_fits and parallel given to Model

  CAHLR/pyBKT#120 · Filter by author ChristianMurphy opened on Oct 29, 2026

• Index the data by student and by skill, and let fit choose skill, student or SSI mode

  CAHLR/pyBKT#119 · Filter by author ChristianMurphy opened on Oct 28, 2026

• Add asv benchmarks, so a PR's effect on speed shows up before it merges

  CAHLR/pyBKT#118 · Filter by author ChristianMurphy opened on Oct 28, 2026

• Type-check pyBKT with mypy and ship type hints

  CAHLR/pyBKT#110 · Filter by author ChristianMurphy opened on Oct 23, 2026

• Lint the C++ and Python code in CI with clang-tidy and Ruff

  CAHLR/pyBKT#109 · Filter by author ChristianMurphy opened on Oct 23, 2026

• Proposal: a logo for pyBKT

  CAHLR/pyBKT#104 · Filter by author ChristianMurphy opened on Oct 23, 2026

• The C++ E-step looks up fixed values under the wrong keys

  CAHLR/pyBKT#99 · Filter by author ChristianMurphy opened on Oct 22, 2026

• Optional SQUAREM acceleration and beta priors for fitting

  CAHLR/pyBKT#85 · Filter by author ChristianMurphy opened on Oct 9, 2026
  2

• The pure Python version runs each expectation step about 550 times slower than C++

  CAHLR/pyBKT#84 · Filter by author ChristianMurphy opened on Oct 9, 2026

• convert_data scans every row once per skill and takes most of Model.fit's time

  CAHLR/pyBKT#83 · Filter by author ChristianMurphy opened on Oct 9, 2026
  3