CHANGELOG.md

Change Log

0.4.9

• Revert back to original version handling. The flipping back and forth was because I had problems with installing the package in NixOS through fetchPypi. More specifically setuptools related checks were failing. Then I realized that I can disable checks.

0.4.8

• Refactor version handling

0.4.7

• Allow failure in importing plot module

0.4.6

Add

• __version__.py for handling versioning

0.4.5

Change

• Readme: back to original form with good examples
• Documentation: Move good parts of new readme to Sphinx documentation

0.4.4

Fix

• Missing images in PyPi

0.4.3

Change

• Absolute to relative image links so that images show in PyPi

0.4.2

Add

• Covariance matrix visualization plot
• Polynomial formula
• Lift for ronia.arraymapper.ArrayMapper

Change

• Design matrix builder function design_matrix to a method of ronia.Formula
• Naming convention: "basis" -> "term"

0.4.1

Fix

• White noise GP
• Remove future annotations which break Py < 3.7 compatibility

0.4.0

Add

• Numpy engine for fitting GAM models
• PyTest configuration file conftest.py
• Documentation with Sphinx
• Unit test coverage

Fix

• Truncated eigenvalue decomposition: np.linalg.eigh -> np.linalg.svd. Eigh is problematic since it may return negative eigenvalues (very close to zero). As a consequence, the truncated covariance decomposition never gives the full basis with sensible arguments. See also https://github.com/roniawz/ronia/issues/5.

0.3.4

Fix

• Bug in setup.py

0.3.3

Remove

• Versioneer
• Redundant information in setup.py

0.3.2

Remove

• Dependence from attrs package

0.3.1

Add

• Common constructor for kernel based GP formulas
• Unit tests for GP kernels

Fix

• Numpy testing incompatibility issues
• Numpy deprecation warning

0.3.0

Add

• Type hints for most functions
• Unit testing module for modules formulae, utils
• Unit tests for BayesianGAM
• Docstrings

0.2.3

Add

• Parameter verbose to BayesianGAM.fit defaulting to False

0.2.2

Add

• Support serializing models to JSON

0.2.1

Add

• Methods predict_marginal and predict_marginals to BayesianGAM
• Type hints to BayesianGAM

0.2.0

Change

• BayesPyFormula => Formula
• Formulae package => module

Fix

• __div__ to __truediv__ in ArrayMapper

Remove

• Broken comparison methods in ArrayMapper

Add

• Unit tests for ArrayMapper

0.1.0

Change

• BayesPyFormula interface: Attribute priors => prior. This gives more flexibility in defining priors. E.g., when summing up terms, sometimes one wants the final prior not to be block diagonal.
• Naming *partial* => *marginal* in BayesianGAM

Add

• Sum and Flatten transformations for BayesPyFormula
• Smoke tests corresponding the README examples

0.0.4

Remove

• Dependence from Pandas

0.0.3

Fix

• Simplify BayesPyFormula.__mul__ so that all bases are just multiplied with the given input map
• All references to sigma now correspond to variance

0.0.2

Add

• Formulas for BSpline1d and ReLU

Change

• KeyFunction is now called ArrayMapper

Remove

• attr frozen class feature from BayesianGAM in order to be able to define theta as the prior distribution by default in the initialization

0.0.1

Add

• Create package basic structure
• BayesPyFormula for configuring and manipulating basis-function based GAM models
• BayesianGAM for fitting predictive GAM models using Bayesian methods
• Utilities for constructing Gaussian process covariances using suitable kernel functions
• kron constructing multi-dimensional Gaussian processes with Kronecker covariance structure
• Various plotting tools for validating and inspecting model parameters and their posterior probability distributions
• Three illustrated examples to README.md

X.Y.Z

Add

Change

Fix

Remove