v0.12.4

DecisionTree v0.12.4

Diff since v0.12.3

• (bugfix) This release fixes a bug where labels would internally always be converted to Float64 (#226). This is now fixed thanks to @xinadi!

Merged pull requests:

• Apply JuliaFormatter (#219) (@rikhuijzer)
• Add .JuliaFormatter.toml (#221) (@rikhuijzer)
• Float64 replaced by AbstractFloat for regression (#226) (@xinadi)
• For a 0.12.4 release (#227) (@rikhuijzer)

Closed issues:

• Multi-variable RandomForestRegressor in Julia (#190)
• Standardize the way fit! and predict methods take X matrix (features) (#220)
• Is out-of-bag error of RandomForestClassifier implementable ? (#223)
• documentation: Clarify n_subfeatures in build_tree? (#224)

v0.12.3

DecisionTree v0.12.3

Diff since v0.12.2

Closed issues:

• Multithreading in apply_forrest (#134)
• Replicate Python model in Julia (#203)
• Fail to precompile the DecisionTree.jl on M1 mac (#212)
• Add functionality for adding trees to an existing forest (#213)

Merged pull requests:

• document use_multithreading keyword (#208) (@rafaqz)
• Add docstring for build_forest (#214) (@ablaom)
• Add method build_forest(model, ...) to add trees to existing model (#216) (@ablaom)

v0.12.2

DecisionTree v0.12.2

Diff since v0.12.1

Closed issues:

• Feature importance from random forest regression (#204)

Merged pull requests:

• Bump AbstractTrees.jl compat for 0.12 series (#205) (@ablaom)
• For a 0.12.2 release (#207) (@ablaom)

v0.11.4

DecisionTree v0.11.4 (RELEASED AFTER 0.12.0)

Diff since v0.11.3

This release bumps the compat requirement for the dependency AbstractArrays.jl which was previously set incorrectly.

v0.12.1

DecisionTree v0.12.1

Diff since v0.12.0

Closed issues:

• Round thresholds in display of trees (#197)

Merged pull requests:

• Improved label printing/plotting of nodes/leaves (#200) (@roland-KA)
• For a 0.12.1 release (#201) (@ablaom)

v0.12.0

DecisionTree v0.12.0

Diff since v0.11.3

• (breaking) For each tree in a random forest models, use seed! to put a copy of each associated RNG into a unique state, to mitigate observed correlation between trees (#194, #198) @dhanak @rikhuijzer

Closed issues:

• Citation / Reference for DecisionTree.jl (#193)
• RNG “shuffling” introduced in #174 is fundamentally flawed (#194)

Merged pull requests:

• Use seed! to put every copy of rng into a unique state (#198) (@dhanak)
• For a 0.12.0 release (#199) (@ablaom)

v0.11.3

DecisionTree v0.11.3

Diff since v0.11.2

• (enhancement) #195 @roland-KA

Merged pull requests:

• Adapt to new AbstractNode in AbstractTrees (#195) (@roland-KA)
• For a 0.11.3 release (#196) (@ablaom)

v0.11.2

DecisionTree v0.11.2

Diff since v0.11.1

• (enhancement) Add option to use multithreading in getting predictions for random forests, as in apply_forest(forest, features; use_multithreading=true). The default continues to be no threading. (#188) @salbert83

Merged pull requests:

• Multithreading for random forest predictions (#188) (@ablaom)
• For a 0.11.2 release (#189) (@ablaom)

v0.11.1

DecisionTree v0.11.1

Diff since v0.11.0

Merged pull requests:

• Fix print_tree (#185) (@rikhuijzer)
• Fix type piracy on zero (#186) (@rikhuijzer)
• For a 0.11.1 release (#187) (@ablaom)

v0.11.0

DecisionTree v0.11.0

Diff since v0.10.13

• Bump minimum Julia version to 1.6.
• (enhancement) Add methods impurity_importance, split_importance and permutation_importance for all models (#182) @yufongpeng
• (breaking) Use the SAMME algorithm for calculating AdaBoostStumpClassifier coefficients (#167) @yufongpeng
• (breaking) Do not export the functions R2, mean_sqaure_error, majority_vote, confusion_matrix and ConfusionMatrix (#183)
• Add accuracy to built-in measures but do not export.

Closed issues:

• [Feature Request] Add a Field for Feature Importance (#170)
• Remove the Int as rng functionality (#177)
• Stop exporting metrics? (#181)

Merged pull requests:

• Adabooststump algorithm change (#167) (@yufongpeng)
• Bump compat julia="1.6", AbstractArrays="0.3,0.4"; adjust CI (#179) (@ablaom)
• Feature importance (#182) (@yufongpeng)
• Stop exporting measures (#183) (@ablaom)
• For a 0.11 release (#184) (@ablaom)