Skip to content

Latest commit

 

History

History
70 lines (56 loc) · 3.13 KB

CHANGELOG.rst

File metadata and controls

70 lines (56 loc) · 3.13 KB
Raw

Release Changelog

0.3.0 (2021-12-09)

  • Allow X to contain NaN values on all estimators except local linear.
  • Allow skgrf predictors to work with shap using skgrf.utils.shap.shap_patch context manager
  • Fix Tree.values for classifiers
  • Fix Tree.feature to use proper value for leaf nodes
  • Fix package includes to prevent installing extra files to site-packages
  • Drop support for Python 3.6

0.2.1 (2021-09-03)

  • Use oldest supported numpy for builds

0.2.0 (2021-08-23)

  • Pin sklearn version to >=0.23.0,<1.0
  • Changed tree details calculations to be optional due to expensive calculations
  • Added more documentation around tree detail interface

0.1.1 (2021-07-20)

  • Fix to include grf src (accidentally removed in 0.1.0)

0.1.0 (2021-07-04)

  • Change the default number of estimators in causal regression forest from 100 to 2000
  • Add validation to ensure the resulting records after taking sample fraction is always at least 1
  • Change n_features_ attribute to n_features_in_ for consistency with sklearn
  • Add validation to sample_weight. Because of internal rng, this ensures that passing no weights is equivalent to passing a vector of ones
  • Add validation to the number of features in the test set.
  • Fitted estimators now hold a pointer to a C++ forest, eliminating the need for deserialization on predict which makes predictions faster
  • Change the attribute name from grf_forest_ to boosted_forests_ on GRFBoostedRegressor
  • Add decision tree regressors for each forest type (except boosted)
  • Add a GRFBaseTree class for shared methods on decision trees
  • Add estimators_ property to forests which return a list of tree estimators
  • Add get_estimator method to forests to extract a single estimator
  • Fix a bug on local linear forest where ll_split_variables_ was not being set properly
  • Add get_feature_importances to forests
  • Add get_split_frequencies to forests
  • Add get_kernel_weights to forests
  • Add COPYING, README.md, REFERENCE.md source files from grf to package source files
  • Remove leading underscore from all grf_forest_ dictionary keys
  • Implement most of the sklearn low-level Tree API into decision trees
  • Add criterion (str) attribute to all estimators
  • Remove arguments related to sparse data from bindings
  • Update grf to 2.0, and update bindings
  • Add GRFClassifier and GRFTreeClassifier estimators
  • Change classes_ and n_classes_ to clusters_ and n_clusters_. classes_ and n_classes_ are now used for target classes in classification estimators
  • Remove the DataNumpy.cpp class. This is no longer needed with the grf 2.0 updates to the Data.cpp class
  • Rename the DataNumpy Cython class to Data
  • Remove extern cpp includes from Cython .pxd file and add cpp files to Extension's source arg
  • Rename forest estimators to GRFForest...

0.0.1 (2021-04-11)

  • Modify serialize_forest function to support linux wheels.

0.0.0 (2021-02-23)

  • First release.