fix: fuse StandardScaler::transform to single-allocation pass (#450)

* fix: fuse StandardScaler::transform to single-allocation pass

Replace per-column take_column/sub_scalar/div_scalar/build_matrix_from_columns
pattern with a single M::fill + element-wise (x - mean) / std loop.

Eliminates O(d) medium Vec allocations and several full-matrix temporaries
that caused ~9500x wall-time regression and RSS inflation on large matrices
(4000x4000: 85.9s → 0.009s per issue #449).

Remove now-dead build_matrix_from_columns helper and its test.
Add comprehensive test covering all parameter combinations (with_mean,
with_std, zero-variance columns, column-count mismatch) verified
against numpy.

* fix(xgboost): harden empty-data guard per #448 review

- Extend guard to also reject zero-feature matrices (n_features == 0)
- Improve error message to 'Training data must contain at least one
  sample and one feature.'
- Add test_fit_on_zero_features_returns_error
- Add comment on scaffold matrix in test_fit_on_empty_data_returns_error

* fix(tree): guard all tree/ensemble fit methods against empty data

Add n_samples == 0 || n_features == 0 guards to:
- DecisionTreeClassifier::fit
- BaseTreeRegressor::fit
- RandomForestClassifier::fit
- BaseForestRegressor::fit

All return FailedError::ParametersError with message:
'Training data must contain at least one sample and one feature.'

Previously these could panic (divide-by-zero, empty-range) or produce
undefined models when called with zero-row or zero-column matrices.

Regression tests added for each guarded path.

* fix(preprocessing): row-major loop order in StandardScaler::transform

Swap fused transform loop from column-outer/row-inner to row-outer/
col-inner with pre-computed (mean, std) Vec. DenseMatrix uses row-major
layout, so the previous ordering caused strided reads and writes on
large matrices.

Also add zero-features regression tests for DecisionTreeClassifier and
BaseForestRegressor to match BaseTreeRegressor coverage.

Audit: ExtraTreesRegressor and RandomForestRegressor both delegate to
BaseForestRegressor::fit which already has the guard — no changes needed.

Addresses review feedback from Mec-iS on PR #450.

Author: Mec-iS

Cargo.toml

@@ -2,7 +2,7 @@
 name = "smartcore"
 description = "Machine Learning in Rust."
 homepage = "https://smartcorelib.github.io/"
-version = "0.6.12"
+version = "0.6.13"
 authors = ["smartcore Developers"]
 edition = "2024"
 rust-version = "1.85"

src/ensemble/base_forest_regressor.rs

@@ -89,6 +89,12 @@ impl<TX: Number + FloatNumber + PartialOrd, TY: Number, X: Array2<TX>, Y: Array1
         if n_rows != y.shape() {
             return Err(Failed::fit("Number of rows in X should = len(y)"));
         }
+        if n_rows == 0 || num_attributes == 0 {
+            return Err(Failed::because(
+                FailedError::ParametersError,
+                "Training data must contain at least one sample and one feature.",
+            ));
+        }
 
         let mtry = parameters
             .m
@@ -223,6 +229,7 @@ impl<TX: Number + FloatNumber + PartialOrd, TY: Number, X: Array2<TX>, Y: Array1
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::linalg::basic::arrays::Array;
     use crate::linalg::basic::matrix::DenseMatrix;
 
     #[test]
@@ -244,4 +251,58 @@ mod tests {
         assert_eq!(regressor.trees.unwrap().len(), 5);
         assert!(regressor.samples.is_some());
     }
+
+    #[test]
+    fn test_fit_on_empty_data_returns_error() {
+        // 2 rows x 2 features — values are arbitrary; only the empty-row case is under test
+        let full = DenseMatrix::from_2d_vec(&vec![vec![1.0, 2.0], vec![3.0, 4.0]]).unwrap();
+        let empty = full.take(&[] as &[usize], 0);
+        assert_eq!(empty.shape(), (0, 2));
+
+        let y: Vec<f64> = vec![];
+        let result = BaseForestRegressor::fit(
+            &empty,
+            &y,
+            BaseForestRegressorParameters {
+                max_depth: None,
+                min_samples_leaf: 1,
+                min_samples_split: 2,
+                n_trees: 5,
+                m: None,
+                keep_samples: false,
+                seed: 0,
+                bootstrap: true,
+                splitter: crate::tree::base_tree_regressor::Splitter::Best,
+            },
+        );
+        assert!(result.is_err());
+        assert_eq!(result.err().unwrap().error(), FailedError::ParametersError);
+    }
+
+    #[test]
+    fn test_fit_on_zero_features_returns_error() {
+        // 2 rows x 2 features — values are arbitrary; only the zero-feature case is under test
+        let full = DenseMatrix::from_2d_vec(&vec![vec![1.0, 2.0], vec![3.0, 4.0]]).unwrap();
+        let no_features = full.take(&[] as &[usize], 1);
+        assert_eq!(no_features.shape(), (2, 0));
+
+        let y: Vec<f64> = vec![1.0, 2.0];
+        let result = BaseForestRegressor::fit(
+            &no_features,
+            &y,
+            BaseForestRegressorParameters {
+                max_depth: None,
+                min_samples_leaf: 1,
+                min_samples_split: 2,
+                n_trees: 5,
+                m: None,
+                keep_samples: false,
+                seed: 0,
+                bootstrap: true,
+                splitter: crate::tree::base_tree_regressor::Splitter::Best,
+            },
+        );
+        assert!(result.is_err());
+        assert_eq!(result.err().unwrap().error(), FailedError::ParametersError);
+    }
 }

src/ensemble/random_forest_classifier.rs

@@ -461,6 +461,12 @@ impl<TX: FloatNumber + PartialOrd, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY
         if x_nrows != y_ncols {
             return Err(Failed::fit("Number of rows in X should = len(y)"));
         }
+        if x_nrows == 0 || num_attributes == 0 {
+            return Err(Failed::because(
+                FailedError::ParametersError,
+                "Training data must contain at least one sample and one feature.",
+            ));
+        }
 
         let mut yi: Vec<usize> = vec![0; y_ncols];
         let classes = y.unique();
@@ -619,6 +625,7 @@ impl<TX: FloatNumber + PartialOrd, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::linalg::basic::arrays::Array;
     use crate::linalg::basic::matrix::DenseMatrix;
     use crate::metrics::*;
 
@@ -779,6 +786,32 @@ mod tests {
         assert!(fail.is_err());
     }
 
+    #[test]
+    fn test_fit_on_empty_data_returns_error() {
+        // 2 rows x 2 features — values are arbitrary; only the empty-row case is under test
+        let full = DenseMatrix::from_2d_vec(&vec![vec![1.0_f64, 1.0], vec![0.0, 1.0]]).unwrap();
+        let empty = full.take(&[] as &[usize], 0);
+        assert_eq!(empty.shape(), (0, 2));
+
+        let y: Vec<u32> = vec![];
+        let result = RandomForestClassifier::fit(
+            &empty,
+            &y,
+            RandomForestClassifierParameters {
+                criterion: SplitCriterion::Gini,
+                max_depth: None,
+                min_samples_leaf: 1,
+                min_samples_split: 2,
+                n_trees: 10,
+                m: None,
+                keep_samples: false,
+                seed: 0,
+            },
+        );
+        assert!(result.is_err());
+        assert_eq!(result.err().unwrap().error(), FailedError::ParametersError);
+    }
+
     #[cfg_attr(
         all(target_arch = "wasm32", not(target_os = "wasi")),
         wasm_bindgen_test::wasm_bindgen_test

src/preprocessing/numerical.rs

@@ -138,76 +138,41 @@ impl<T: Number + RealNumber, M: Array2<T>> UnsupervisedEstimator<M, StandardScal
 /// standard deviation to one.
 impl<T: Number + RealNumber, M: Array2<T>> Transformer<M> for StandardScaler<T> {
     fn transform(&self, x: &M) -> Result<M, Failed> {
-        let (_, n_cols) = x.shape();
-        if n_cols != self.means.len() {
+        let (nrows, ncols) = x.shape();
+        if ncols != self.means.len() {
             return Err(Failed::because(
                 FailedError::TransformFailed,
                 &format!(
                     "Expected {} columns, but got {} columns instead.",
                     self.means.len(),
-                    n_cols,
+                    ncols,
                 ),
             ));
         }
 
-        Ok(build_matrix_from_columns(
-            self.means
-                .iter()
-                .zip(self.stds.iter())
-                .enumerate()
-                .map(|(column_index, (column_mean, column_std))| {
-                    x.take_column(column_index)
-                        .sub_scalar(T::from(self.adjust_column_mean(*column_mean)).unwrap())
-                        .div_scalar(T::from(self.adjust_column_std(*column_std)).unwrap())
-                })
-                .collect(),
-        )
-        .unwrap())
-    }
-}
-
-/// From a collection of matrices, that contain columns, construct
-/// a matrix by stacking the columns horizontally.
-fn build_matrix_from_columns<T, M>(columns: Vec<M>) -> Option<M>
-where
-    T: Number + RealNumber,
-    M: Array2<T>,
-{
-    columns.first().cloned().map(|output_matrix| {
-        columns
-            .iter()
-            .skip(1)
-            .fold(output_matrix, |current_matrix, new_colum| {
-                current_matrix.h_stack(new_colum)
+        let mut output = M::fill(nrows, ncols, T::zero());
+        let col_params: Vec<(T, T)> = (0..ncols)
+            .map(|j| {
+                let mean = T::from(self.adjust_column_mean(self.means[j])).unwrap();
+                let std = T::from(self.adjust_column_std(self.stds[j])).unwrap();
+                (mean, std)
             })
-    })
+            .collect();
+        for i in 0..nrows {
+            for (j, &(mean, std)) in col_params.iter().enumerate() {
+                let val = *x.get((i, j));
+                output.set((i, j), (val - mean) / std);
+            }
+        }
+        Ok(output)
+    }
 }
 
 #[cfg(test)]
 mod tests {
 
     mod helper_functionality {
-        use super::super::{build_matrix_from_columns, ensure_std_valid};
-        use crate::linalg::basic::matrix::DenseMatrix;
-
-        #[test]
-        fn combine_three_columns() {
-            assert_eq!(
-                build_matrix_from_columns(vec![
-                    DenseMatrix::from_2d_vec(&vec![vec![1.0], vec![1.0], vec![1.0],]).unwrap(),
-                    DenseMatrix::from_2d_vec(&vec![vec![2.0], vec![2.0], vec![2.0],]).unwrap(),
-                    DenseMatrix::from_2d_vec(&vec![vec![3.0], vec![3.0], vec![3.0],]).unwrap()
-                ]),
-                Some(
-                    DenseMatrix::from_2d_vec(&vec![
-                        vec![1.0, 2.0, 3.0],
-                        vec![1.0, 2.0, 3.0],
-                        vec![1.0, 2.0, 3.0]
-                    ])
-                    .unwrap()
-                )
-            )
-        }
+        use super::super::ensure_std_valid;
 
         #[test]
         fn negative_value_should_be_replace_with_minimal_positive_value() {
@@ -426,6 +391,128 @@ mod tests {
             )
         }
 
+        /// Verify transform correctness across all parameter combinations.
+        #[test]
+        fn transform_all_parameter_combinations() {
+            let data = DenseMatrix::from_2d_vec(&vec![
+                vec![1.0, 10.0, 100.0],
+                vec![2.0, 20.0, 200.0],
+                vec![3.0, 30.0, 300.0],
+                vec![4.0, 40.0, 400.0],
+            ])
+            .unwrap();
+
+            // Default: with_mean=true, with_std=true
+            // std = population std: sqrt(mean of squared deviations)
+            // For [1,2,3,4]: mean=2.5, pop_std = sqrt(5/4) ≈ 1.1180339887
+            let scaler = StandardScaler::fit(&data, StandardScalerParameters::default()).unwrap();
+            let result = scaler.transform(&data).unwrap();
+            let expected = DenseMatrix::from_2d_vec(&vec![
+                vec![
+                    -1.3416407864998738,
+                    -1.3416407864998738,
+                    -1.3416407864998738,
+                ],
+                vec![
+                    -0.4472135954999579,
+                    -0.4472135954999579,
+                    -0.4472135954999579,
+                ],
+                vec![0.4472135954999579, 0.4472135954999579, 0.4472135954999579],
+                vec![1.3416407864998738, 1.3416407864998738, 1.3416407864998738],
+            ])
+            .unwrap();
+            assert!(
+                result.approximate_eq(&expected, 1e-10),
+                "Default transform failed:\n{result}\nexpected:\n{expected}"
+            );
+
+            // with_mean=true, with_std=false
+            let scaler = StandardScaler::fit(
+                &data,
+                StandardScalerParameters {
+                    with_mean: true,
+                    with_std: false,
+                },
+            )
+            .unwrap();
+            let result = scaler.transform(&data).unwrap();
+            let expected = DenseMatrix::from_2d_vec(&vec![
+                vec![-1.5, -15.0, -150.0],
+                vec![-0.5, -5.0, -50.0],
+                vec![0.5, 5.0, 50.0],
+                vec![1.5, 15.0, 150.0],
+            ])
+            .unwrap();
+            assert!(
+                result.approximate_eq(&expected, 1e-10),
+                "with_mean=true, with_std=false transform failed:\n{result}\nexpected:\n{expected}"
+            );
+
+            // with_mean=false, with_std=true: (x - 0) / std = x / std
+            let scaler = StandardScaler::fit(
+                &data,
+                StandardScalerParameters {
+                    with_mean: false,
+                    with_std: true,
+                },
+            )
+            .unwrap();
+            let result = scaler.transform(&data).unwrap();
+            let expected = DenseMatrix::from_2d_vec(&vec![
+                vec![0.8944271909999159, 0.8944271909999159, 0.8944271909999159],
+                vec![1.7888543819998317, 1.7888543819998317, 1.7888543819998317],
+                vec![2.6832815729997477, 2.6832815729997477, 2.6832815729997477],
+                vec![3.5777087639996634, 3.5777087639996634, 3.5777087639996634],
+            ])
+            .unwrap();
+            assert!(
+                result.approximate_eq(&expected, 1e-10),
+                "with_mean=false, with_std=true transform failed:\n{result}\nexpected:\n{expected}"
+            );
+
+            // with_mean=false, with_std=false (passthrough)
+            let scaler = StandardScaler::fit(
+                &data,
+                StandardScalerParameters {
+                    with_mean: false,
+                    with_std: false,
+                },
+            )
+            .unwrap();
+            let result = scaler.transform(&data).unwrap();
+            assert!(
+                result.approximate_eq(&data, 1e-10),
+                "with_mean=false, with_std=false should return data unchanged:\n{result}\nexpected:\n{data}"
+            );
+
+            // Zero-variance column mixed with normal columns
+            let mixed = DenseMatrix::from_2d_vec(&vec![
+                vec![1.0, 5.0],
+                vec![2.0, 5.0],
+                vec![3.0, 5.0],
+                vec![4.0, 5.0],
+            ])
+            .unwrap();
+            let scaler = StandardScaler::fit(&mixed, StandardScalerParameters::default()).unwrap();
+            let result = scaler.transform(&mixed).unwrap();
+            let expected = DenseMatrix::from_2d_vec(&vec![
+                vec![-1.3416407864998738, 0.0],
+                vec![-0.4472135954999579, 0.0],
+                vec![0.4472135954999579, 0.0],
+                vec![1.3416407864998738, 0.0],
+            ])
+            .unwrap();
+            assert!(
+                result.approximate_eq(&expected, 1e-10),
+                "Zero-variance mixed column transform failed:\n{result}\nexpected:\n{expected}"
+            );
+
+            // Column count mismatch returns error: scaler expects 2 cols, data has 1
+            let narrow = DenseMatrix::from_2d_vec(&vec![vec![1.0]]).unwrap();
+            assert!(scaler.transform(&narrow).is_err());
+        }
+
         /// Same as `fit_for_random_values` test, but using a `StandardScaler` that has been
         /// serialized and deserialized.
         #[cfg_attr(