Skip to content

Latest commit

 

History

History
executable file
·
130 lines (102 loc) · 4.3 KB

CHANGES.rst

File metadata and controls

executable file
·
130 lines (102 loc) · 4.3 KB

Release Notes

HDDM 0.5.5 (bugfix release)

  • Upgrade dependency to pymc 2.3.3
  • Remove LBA model as likelihood seems broken

HDDM 0.5.3 (bugfix release)

  • Compatibility with pandas > 0.13.
  • Fix problem that causes stats to not be generated when loading model.
  • Update packages to work with anaconda 1.9.

HDDM 0.5.2 (bugfix release)

  • Refactored posterior predictive plots and added tutorial: http://ski.clps.brown.edu/hddm_docs/tutorial_post_pred.html
  • Smaller bugfixes.
  • Works with PyMC 2.3.
  • Experimental features:
    • Updated HLBA model but currently has bad recovery.
    • Added sample_emcee() to use the emcee parallel sampler. Seems to work but requires some tuning and does not seem to beat slice sampling.

HDDM 0.5

HDDM 0.4.1

  • Models are now pickable. (This means they can be loaded and saved. Critically, it is now also trivial to run multiple models in parallel that way.)

HDDM 0.4

License

HDDM 0.4 is now distributed under the simplified BSD license (see the LICENSE file) instead of GPLv3.

New features

Bugfixes

  • model.load_db() is working again.

HDDM 0.3.1

  • Fixed annoying bug that broke plotting of posterior predictive.

HDDM 0.3 (6 Sep 2012)

  • Complete rewrite of the underlying model creation engine (kabuki) to allow for more flexible model creation including transforms. This enabled development of a new HDDM default model without explicit parameter bounds.
  • Group mean distributions are now Gibbs sampled and group variability distributions are now slice sampled leading to much improved convergence and mixing.
  • MAP approximation of hierarchical models for better initialization.
  • Improved documentation (check out the `How-to`_ section).
  • Chi-square fitting using the Ratcliff quantile method.
  • Posterior predictive checks.

HDDM 0.2 (First publicly announced release)

  • Better model initialization that shouldn't fail.
  • Many bugfixes.
  • Major internal overhaul.

HDDM 0.1 (20 July 2011) (Semi-private MathPsych release)

  • Flexible HDDM model class to fit group and subject models.
  • Heavily optimized cython likelihoods.