NOCATS: Categorical splits for tree-based learners by jblackburne · Pull Request #4899 · scikit-learn/scikit-learn

Remove inverted dependence of _utils.pyx on _tree.pyx because it was causing all kinds of problems. Now safe_realloc requires the item size to be explicitly provided. Also, it can allocate arrays o…

jblackburne

committed

Tree constructor now checks for mismatched struct sizes.

jblackburne

committed

Created SplitValue datatype to generalize the concept of a threshold to categorical variables. Replaced the threshold attribute of SplitRecord and Node with SplitValue.

jblackburne

committed

Added attribute n_categories to Splitter and Tree, an array of ints that defaults to -1 for each feature (indicating non-categorical).

jblackburne

committed

Added a goes_left function to replace threshold comparisons during prediction with trees. Also introduced category caches for quick evaluation of categorical splits.

jblackburne

committed

BestSplitter now calculates the best categorical split.

jblackburne

committed

Added categorical split code to RandomSplitter.node_split

jblackburne

committed

Added an implementation of the Breiman sorting shortcut for finding the best categorical split.

jblackburne

committed

Added categorical constructor parameter and error checking to BaseDecisionTree.

jblackburne

committed

Added the categorical keyword to forest constructors.

jblackburne

committed

Added some unit tests.

jblackburne

committed

Refactored _partial_dependence_tree a little.

jblackburne

committed

Added categorical support to gradient boosting.

jblackburne

committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

NOCATS: Categorical splits for tree-based learners#4899

NOCATS: Categorical splits for tree-based learners#4899
jblackburne wants to merge 13 commits intoscikit-learn:masterfrom
jblackburne:NOCATS

Commits on Feb 11, 2017

Remove inverted dependence of _utils.pyx on _tree.pyx because it was causing all kinds of problems. Now safe_realloc requires the item size to be explicitly provided. Also, it can allocate arrays o…

Tree constructor now checks for mismatched struct sizes.

Created SplitValue datatype to generalize the concept of a threshold to categorical variables. Replaced the threshold attribute of SplitRecord and Node with SplitValue.

Added attribute n_categories to Splitter and Tree, an array of ints that defaults to -1 for each feature (indicating non-categorical).

Added a goes_left function to replace threshold comparisons during prediction with trees. Also introduced category caches for quick evaluation of categorical splits.

BestSplitter now calculates the best categorical split.

Added categorical split code to RandomSplitter.node_split

Added an implementation of the Breiman sorting shortcut for finding the best categorical split.

Added categorical constructor parameter and error checking to BaseDecisionTree.

Added the categorical keyword to forest constructors.

Added some unit tests.

Refactored _partial_dependence_tree a little.

Added categorical support to gradient boosting.

Uh oh!

NOCATS: Categorical splits for tree-based learners#4899jblackburne wants to merge 13 commits intoscikit-learn:masterscikit-learn/scikit-learn:masterfrom jblackburne:NOCATSjblackburne/scikit-learn:NOCATSCopy head branch name to clipboard

Commits

Commits on Feb 11, 2017

NOCATS: Categorical splits for tree-based learners#4899
jblackburne wants to merge 13 commits intoscikit-learn:masterfrom
jblackburne:NOCATS