test(tree): pin min_samples_split boundary and assert depth after #464

CHANGELOG.md

@@ -4,6 +4,10 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.6.15]
+### Fixed
+- `tree`: regression tests for the tree growth fix from #464. New `min_samples_split_boundary` tests pin the split rule for `DecisionTreeClassifier` and `BaseTreeRegressor`: a node that holds exactly `min_samples_split` samples must still split, and a node with fewer samples must stay a leaf. The `full_depth` tests now also assert the tree `depth` (3), which guards the public `DecisionTreeClassifier::depth` accessor against silent regressions. Library code is unchanged.
+
 ## [0.6.14]
 ### Added
 - `decomposition/lda.rs`: `LDA`, linear discriminant analysis for supervised dimensionality reduction (#136). It projects the data onto the directions that best separate the classes, keeping `min(n_classes - 1, n_features)` components by default, and implements the `Transformer` interface next to `PCA`. Directions match scikit-learn's `LinearDiscriminantAnalysis(solver="eigen")` up to sign.

Cargo.toml

@@ -2,7 +2,7 @@
 name = "smartcore"
 description = "Machine Learning in Rust."
 homepage = "https://smartcorelib.github.io/"
-version = "0.6.14"
+version = "0.6.15"
 authors = ["smartcore Developers"]
 edition = "2024"
 rust-version = "1.85"

src/tree/base_tree_regressor.rs

@@ -622,6 +622,39 @@ mod tests {
         let y_expected = vec![1.0, 2.0, 6.5, 6.5, 11.50, 11.50];
         let y_hat = tree.predict(&x).expect("Predict should work");
         assert_eq!(tree.nodes().len(), 7);
+        assert_eq!(tree.depth, 3);
         assert!(mean_absolute_error(&y_expected, &y_hat) < 1e-9);
     }
+
+    #[test]
+    fn min_samples_split_boundary() {
+        // A node that holds exactly `min_samples_split` samples must still split.
+        let x = DenseMatrix::from_2d_vec(&vec![vec![1.0_f64], vec![2.0], vec![3.0]]).unwrap();
+        let y = vec![1.0f64, 2.0, 6.0];
+
+        let parameters = BaseTreeRegressorParameters {
+            max_depth: None,
+            min_samples_leaf: 1,
+            min_samples_split: 3,
+            seed: None,
+            splitter: Splitter::Best,
+        };
+
+        let tree = BaseTreeRegressor::fit(&x, &y, parameters).expect("Fit should work");
+        assert_eq!(tree.nodes().len(), 3);
+        assert_eq!(tree.depth, 2);
+
+        // A node with fewer than `min_samples_split` samples must stay a leaf.
+        let parameters = BaseTreeRegressorParameters {
+            max_depth: None,
+            min_samples_leaf: 1,
+            min_samples_split: 4,
+            seed: None,
+            splitter: Splitter::Best,
+        };
+
+        let tree = BaseTreeRegressor::fit(&x, &y, parameters).expect("Fit should work");
+        assert_eq!(tree.nodes().len(), 1);
+        assert_eq!(tree.depth, 0);
+    }
 }

src/tree/decision_tree_classifier.rs

@@ -1078,11 +1078,44 @@ mod tests {
         let tree = DecisionTreeClassifier::fit(&x, &y, parameters).expect("Fit should work");
         let y_hat = tree.predict(&x).expect("Predict should work");
         assert_eq!(tree.nodes().len(), 7);
+        assert_eq!(tree.depth(), 3);
 
         // Tree should have 5 out of 6 examples correct
         assert!((accuracy(&y, &y_hat) - 5.0 / 6.0).abs() < 1e-9);
     }
 
+    #[test]
+    fn min_samples_split_boundary() {
+        // A node that holds exactly `min_samples_split` samples must still split.
+        let x = DenseMatrix::from_2d_vec(&vec![vec![1.0_f64], vec![2.0], vec![3.0]]).unwrap();
+        let y = vec![0usize, 0, 1];
+
+        let parameters = DecisionTreeClassifierParameters {
+            max_depth: None,
+            min_samples_leaf: 1,
+            min_samples_split: 3,
+            seed: None,
+            criterion: SplitCriterion::Gini,
+        };
+
+        let tree = DecisionTreeClassifier::fit(&x, &y, parameters).expect("Fit should work");
+        assert_eq!(tree.nodes().len(), 3);
+        assert_eq!(tree.depth(), 2);
+
+        // A node with fewer than `min_samples_split` samples must stay a leaf.
+        let parameters = DecisionTreeClassifierParameters {
+            max_depth: None,
+            min_samples_leaf: 1,
+            min_samples_split: 4,
+            seed: None,
+            criterion: SplitCriterion::Gini,
+        };
+
+        let tree = DecisionTreeClassifier::fit(&x, &y, parameters).expect("Fit should work");
+        assert_eq!(tree.nodes().len(), 1);
+        assert_eq!(tree.depth(), 0);
+    }
+
     #[cfg_attr(
         all(target_arch = "wasm32", not(target_os = "wasi")),
         wasm_bindgen_test::wasm_bindgen_test