Skip to content

v1.6.0 — Point-process and Kalman-filter EM repairs

Latest
Latest

Choose a tag to compare

View all tags
@iahncajigas iahncajigas released this 06 Oct 21:35
· 15 commits to master since this release
v1.6.0
3f40d97
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Install by double-clicking nSTAT-1.6.0.mltbx in MATLAB, or run matlab.addons.toolbox.installToolbox('nSTAT-1.6.0.mltbx'). Tested on R2026a.

Point-process and Kalman-filter EM release. Before this release, PP_EM could not run in any configuration, and KF_EM's main loop was unreachable. Both families now run and are numerically correct, and the standard-error routines no longer hang on a singular observed information (#135, #137, #138). nstat-python v0.6.0 mirrors these fixes.

Breaking changes

  • EM defaults: the NewtonRaphson M-step is now the default, and x0/Px0 are no longer estimated.
  • Changed outputs: the SE and p-value (Pvals) layout and values change, and the information criteria are now on the original scale. EM now estimates from a non-diagonal Q0/R0 instead of returning the starting values.
  • GLM M-step: choosing MstepMethod='GLM' emits a warning (nSTAT:EM:glmPlugIn).
  • KF_EM: its internal static method is renamed KF_RunEM. The public DecodingAlgorithms.KF_EM entry point is unchanged.

The full per-commit tables are in RELEASE_NOTES.md.

🤖 Generated with Claude Code

Assets 3
Loading