v0.3.3

LRMoE v0.3.3

Diff since v0.3.2

Closed issues:

• Make hyperparameters usage of LogNormal consistent (#31)
• Potential improvement on exposurizing models (#33)
• Improving rowlogsumexp (#34)
• Add print step frequency (#35)
• Further fix numerical integration of logY_sq for LogNormalExpert (#36)
• Consider alternative regularization on LogNormalExpert (#42)
• Improvement on exposurized models (#43)
• Parallelize numerical integration with @threads (#47)

Merged pull requests:

• correct for constant in integration (#37) (@sparktseung)
• fix performence or rowlogsumexp (#38) (@sparktseung)
• fix hyperparameters (#39) (@sparktseung)
• improve print_step (#40) (@sparktseung)
• update lognormal regularization (#44) (@sparktseung)
• Feature/improve exposurize (#45) (@sparktseung)
• Feature/parallelize numerical integration (#48) (@sparktseung)
• remove unnecessary function exports (#50) (@sparktseung)
• Dev 0.3.3 (#51) (@sparktseung)

v0.3.2

LRMoE v0.3.2

Diff since v0.3.1

Bug fix:

• Error of spurious multiplicative constant when integrating logY for lognormal. #24

Internal:

• (Sort of) properly set up CI for package testing #26
• Added a few initial tests for basic functions in the package #26
• Small cleanup of documentations #27

Closed issues:

• Delete obsolete tests and restructure tests (#18)
• Numerical Integration error for logY in LogNormalExpert (#24)

Merged pull requests:

• fix spurious multiplier in integration (#25) (@sparktseung)
• Fix/update tests (#26) (@sparktseung)
• remove unnecessary file (#27) (@sparktseung)
• Dev 0.3.2 (#28) (@sparktseung)

v0.3.1

Major:

• Bump up dependency versions: #14

Bug fix:

• Fix initialization of ZINB: #15
• Fix spurious lognormal fit with small variance: #17

Internal:

• Code format for the entire repo for better maintenance: #16

v0.3.0

Major update: Fitting function with exposure when modelling count data. Now the model can fit inexact count data observed under different time lengths.
Minor updates: bug fixes such as avoiding spurious models with there is no penalty on the expert function by default.

v0.2.0

Major update: add exposure in modelling count data. Now the model can deal with cases where count data are observed over different time intervals for individual observations.
Minor updates: documentation, predictive functions (also with exposure), bug fixes.

v0.1.1

Fix errors in expert_ll/tn/tn_bar when there is an expert parameter named p.

First Release

First release with basic functionalities.