Skip to content

Stabilize discrete-state prior normalization - #4286

Closed
FlorianPfaff wants to merge 17 commits into
mainfrom
agent/stabilize-discrete-state-prior-normalization
Closed

Stabilize discrete-state prior normalization#4286
FlorianPfaff wants to merge 17 commits into
mainfrom
agent/stabilize-discrete-state-prior-normalization

Conversation

@FlorianPfaff

Copy link
Copy Markdown
Owner

Summary

  • normalize finite discrete-state prior vectors after scaling by their maximum entry
  • prevent overflow in the probability sum while preserving relative mass
  • add a public forward/backward regression with extreme finite initial probabilities

Bug

The compatibility-layer probability validator normalized priors with values / values.sum(). A valid vector can contain individually finite values whose sum overflows to infinity. For example, [finfo.max, finfo.max / 2] has a mathematically well-defined normalized ratio of 2:1, but the direct floating-point sum is inf, so the implementation returned [0, 0]. The forward recursion then reported that the first emission had no prior-supported likelihood mass.

Fix

Scale the non-negative vector by its largest entry before summing. All scaled entries are in [0, 1], the relative probabilities are unchanged, and the positive-mass check remains explicit.

Validation

  • reproduced the pre-fix overflow and zero-vector normalization with NumPy
  • verified scale-first normalization yields [2/3, 1/3]
  • added tests/filters/test_discrete_state_probability_overflow.py
  • branch comparison against main: two commits ahead, zero behind; 7 production-line changes and one focused regression file

The full repository test matrix is delegated to GitHub Actions.

@github-actions

github-actions Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor

MegaLinter analysis: Success

Descriptor Linter Files Fixed Errors Warnings Elapsed time
✅ COPYPASTE jscpd yes no no 24.92s
✅ JSON prettier 7 0 0 0 0.92s
✅ JSON v8r 7 0 0 3.59s
✅ MARKDOWN markdownlint 68 0 0 0 1.81s
✅ MARKDOWN markdown-table-formatter 68 0 0 0 0.68s
✅ PYTHON black 1490 182 0 0 95.51s
✅ PYTHON isort 1490 327 0 0 2.65s
✅ REPOSITORY betterleaks yes no no 2.26s
✅ REPOSITORY checkov yes no no 53.9s
✅ REPOSITORY gitleaks yes no no 14.85s
✅ REPOSITORY git_diff yes no no 0.44s
✅ REPOSITORY secretlint yes no no 60.91s
✅ REPOSITORY syft yes no no 5.08s
✅ REPOSITORY trivy-sbom yes no no 6.28s
✅ REPOSITORY trufflehog yes no no 31.84s
✅ YAML prettier 11 0 0 0 0.78s
✅ YAML v8r 11 0 0 11.6s
✅ YAML yamllint 11 0 0 0.62s

Notices

📣 MegaLinter 9.5.0 is out! Discover the new features and security recommendations in the release announcement. (Skip this info by defining SECURITY_SUGGESTIONS: false)

See detailed reports in MegaLinter artifacts

Your project could benefit from a custom flavor, which would allow you to run only the linters you need, and thus improve runtime performances. (Skip this info by defining FLAVOR_SUGGESTIONS: false)

  • Documentation: Custom Flavors
  • Command: npx mega-linter-runner@9.6.0 --custom-flavor-setup --custom-flavor-linters PYTHON_BLACK,PYTHON_ISORT,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_GITLEAKS,REPOSITORY_BETTERLEAKS,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,REPOSITORY_TRUFFLEHOG,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

MegaLinter is graciously provided by OX Security
Show us your support by starring ⭐ the repository

@FlorianPfaff
FlorianPfaff force-pushed the agent/stabilize-discrete-state-prior-normalization branch from f743a93 to f06a95e Compare July 13, 2026 16:36

Copy link
Copy Markdown
Owner Author

Closing as superseded by #4405, the newer current-main implementation of stable discrete-state prior normalization.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant