feat: random improvements 2025 08 17

[eschmidt42]

This pull request refactors the codebase to move all decision tree–related modules into a new models subpackage, updates all import statements accordingly, and introduces utility functions to handle random sampling of features and samples. It also includes some minor improvements to scikit-learn compatibility and code clarity. The changes affect both the library code and the example notebooks.

1. Codebase Restructuring and Import Updates

• Moved all decision tree modules (e.g., estimators.py, node.py, predict.py, split.py, train.py, visualize.py, and related objects) from random_tree_models/decisiontree/ to random_tree_models/models/decisiontree/ and updated all corresponding import statements throughout the codebase and notebooks. [1] [2] [3] [4] [5] [6] [7] F011abadL3R3, [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21]

2. Utility Functions for Random Sampling

• Added new utility functions get_random_sample_ids and get_random_feature_ids in random_tree_models/models/decisiontree/random.py to encapsulate logic for randomly selecting samples and features, and refactored their usage in the decision tree estimator code. [1] [2]

3. Improvements to scikit-learn Compatibility

• Updated the way scikit-learn estimator tags are handled for decision tree classifiers to use ClassifierTags and the __sklearn_tags__ method, improving integration with scikit-learn's estimator checks. [1] [2]

4. Minor Enhancements and Cleanups

• Replaced usage of the old dtree alias with direct imports of DecisionTreeClassifier and DecisionTreeRegressor in ensemble models (ExtraTrees, GradientBoostedTrees), and simplified type annotations. (F011abadL3R3, [1] [2] [3] [4] [5] [6]
• Added a check to ensure n_trees is greater than zero in GradientBoostedTreesRegressor using is_greater_zero.

These changes modernize the code structure, improve maintainability, and enhance compatibility with scikit-learn conventions.