SDT.jl

Appearance

Changelog ​

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

v1.4.0 ​

  • added support Bagging models as a component in an Ensemble

  • added support for null classifiers based on Bagging models

  • added variable selection for Bagging models #405

  • added a more general syntax for variable selection (VariableSelectionStrategy)

  • added a ChainedTransform type to chain two data transformation steps, useful for PCA + z-score #408

  • fixed the issue with variable selection reseting the model variables #400

  • added a method for noselection! where no folds are given

v1.3.4 ​

  • improved the memory allocation of Logistic regression

v1.3.3 ​

  • fixed the problem with hyper-parameters not being restored #371

  • fixed the typing issue in Shapley values #392

v1.3.2 ​

  • fixed a bug where the Shapley values calculation would not work when applied to a dataset not used for testing #375

v1.3.1 ​

  • fixed a bug where the verbose output of variable selection was wrong #372

v1.3.0 ​

  • added the option to report on validation loss when training a logistic #365

  • added the option to pass arbitrary keywords to the training of the classifier

  • improved the verbose output of logistic regression #364

  • improved the internals of train! to use a simpler syntax

  • improved the verbose output of variable selection #366

v1.2.3 ​

  • fixed the threading mechanism for cross-validation #458

  • added a QOL function to sum confusion matrices

v1.2.2 ​

  • added the verbose field to Logistic to determine whether to print an output for gradient descent

  • added the interactions field to Logistic to determine which interactions to include

  • changed the creation of parameters for Logistic to allow only some interaction terms

  • improved the performance (speed and memory consumption) of training logistic regressions

v1.2.1 ​

  • added Logistic as a new classifier using logistic regression with gradient descent

v1.2.0 ​

  • changed the default for bagging to maintain class balance in all bagged models

v1.1.2 ​

  • improved the performance (speed and memory requirement) of prediction with Naive Bayes

  • improved the performance (speed) of training BIOCLIM

  • improved the performance (GC) of Shapley explanations

  • improved the performance (speed) of variable importance

v1.1.1 ​

  • improved the performance (speed and memory requirement) of training decision trees

v1.1.0 ​

  • added backward selection with protected variables (to mirror forced variables in forward selection)

  • added stratification of presence/absence for cross-validation (prevalence is maintained across folds)

v1.0.0 ​

  • changed the default training option for transformers to be presence-only, with the absences=true keyword to use absences as well

  • added the transformer and classifier methods, that return the transformer and classifier of SDM and Bagging models