Skip to content

Latest commit

 

History

History
148 lines (114 loc) · 5.27 KB

NEWS.md

File metadata and controls

148 lines (114 loc) · 5.27 KB
Raw

BranchGLM News

BranchGLM 2.1.6

  • Adding plot method for BranchGLM objects
  • Fixing another very small bug in branch and bound algorithms
  • Fixing bug in log-likelihood calculation for gaussian models when the number of observations is odd

BranchGLM 2.1.5

  • Fixing a small bug with parallel computation for forward and switch branch and bound algorithms.
  • Modifying stepwise variable selection algorithms to return the whole solution path which can now be accessed with predict() or coef() and can be plotted with plot()

BranchGLM 2.1.4

  • Updating all documentation and adding more tests
  • Fixing a couple of bugs

BranchGLM 2.1.3

  • Updating predict functions to include na.action and to better handle offset variables.
  • The VariableSelection() function now returns the coefficients as part of the selection process, so fit() doesn’t need to be used for every call to coef() or predict().
  • Adding new cols arguments to plot.summary.BranchGLMVS() and plot.BranchGLMVS() to allow for control over the colors

BranchGLM 2.1.2

  • Adding new cex arguments to plot.summary.BranchGLMVS() and plot.BranchGLMVS() to allow for more control over text size
  • Fixing bug in the “variables” plot from plot.summary.BranchGLMVS() that resulted in one of the variables being removed
  • Fixing bug in root-finding algorithm for finding confidence intervals with confint.BranchGLM()

BranchGLM 2.1.1

  • Adding multiple new arguments to plot.summary.BranchGLMVS() and plot.BranchGLMVS()
  • Changing estimate of dispersion for gaussian GLMs to be the MLE instead of the unbiased estimator
  • Adding keepintercept argument to VariableSelection() to specify whether the intercept should be kept or not
  • Removing standard errors and p-values from output of fit() since these are biased due to the variable selection process

BranchGLM 2.1.0

  • Fixing bug in VariableSelection() when using formulas of the form y ~ . - variable
  • Changing default type for VariableSelection() to be “branch and bound”
  • Adding vcov and confint methods for BranchGLM objects
  • Adding plot method for objects resulting from confint.BranchGLM()
  • Adding coef and predict methods for BranchGLMVS objects
  • Fixing bug that caused the switch branch and bound type in VariableSelection() to be slower

BranchGLM 2.0.1

  • Fixing bug in VariableSelection() when using the switch branch and bound method where duplicate models are returned
  • Fixing bug in VariableSelection() when using the switch branch and bound method or branch and bound method where factor variables were handled incorrectly
  • No longer allowing models that failed to converge to be one of the top models in the VariableSelection() function

BranchGLM 2.0.0

  • Added the following features to the VariableSelection() function
    • Finding the top k models according to the metric via the bestmodels argument
    • Finding all models with a metric value within k of the best metric value via the cutoff argument
    • Added HQIC as a possible metric, HQIC is the Hannan-Quinn information criterion
  • Added summary method for BranchGLMVS objects along with the following functions
    • plot.summary.BranchGLMVS() for plotting results from variable selection
    • fit.summary.BranchGLMVS() which can be used to get a BranchGLM object for one of the best models found

BranchGLM 1.3.2

  • Improving efficiency for the “branch and bound” and “switch branch and bound” methods in the VariableSelection() function.
  • Fixed bug related to initial values in the VariableSelection() function.

BranchGLM 1.3.1

  • The VariableSelection() function should now properly handle interaction terms.

BranchGLM 1.3.0

  • Updated GLM fitting to use backtracking line search with strong Wolfe conditions instead of Armijo-Goldstein condition to find step size.
  • Adding new variable selection types for VariableSelection() which are called “backward branch and bound” and “switch branch and bound”. These methods are similar to the regular branch and bound method, but sometimes they can be much faster.
  • Added predict method for BranchGLMVS objects.

BranchGLM 1.2.0

  • Introducing new function BranchGLM.fit() which is similar to glm.fit(), it fits GLMs when given the design matrix X and the outcome vector Y. Can be faster than calling BranchGLM if X and Y are readily available.
  • Fixing number of models fit that are reported by stepwise selection procedures when using parallel computation via the VariableSelection() function.
  • Fixing SEs and p-values for gaussian and gamma GLMs.

BranchGLM 1.1.3

  • Fixing number of observations returned from VariableSelection() function in the presence of missing values

BranchGLM 1.1.2

  • Fixing multiple different bugs
  • Speeding up linear regression fitting, especially for large models

BranchGLM 1.1.1

  • Fixing multiple different bugs

BranchGLM 1.1.0

  • Adding NEWS.md
  • Minimized repeated work for linear regression, so it should now be much faster
  • Added gamma regression along with some additional link functions
  • Additional arguments added to BranchGLM() to reduce memory usage if desired
  • Fixed print statement for BranchGLMVS objects when parallel computation was employed