feat(deep_causality_algorithms): Parallelize mRMR feature selection algo.

This commit introduces parallel execution to the mRMR feature selection algorithm
  in the deep_causality_algorithms crate, significantly improving its performance on
   large datasets.

  The core changes include:
   - Parallelizing the initial and iterative feature selection loops in both
     mrmr_features_selector and mrmr_features_selector_cdl using the rayon crate.
   - Guarding all parallel logic with the existing parallel feature flag, ensuring
     sequential execution remains the default.
   - Correcting bugs in the test suite and benchmark data generation to ensure
     robust validation.
   - Updating documentation to reflect the new parallel execution capability and
     its performance benefits.

  The parallel implementation has been verified to be functionally equivalent to the sequential version and demonstrates a significant speedup, as validated by both synthetic benchmarks and a real-world case study on the ICU sepsis dataset.

Signed-off-by: Marvin Hansen <[email protected]>

Author: marvin-hansen

Cargo.lock

@@ -52,8 +52,7 @@ reveal the nature of multi-variable interactions.
 * **Performance Optimized**:
     * **Algorithmic Capping**: Use the `MaxOrder` enum to limit the analysis to a tractable number of interactions (
       e.g., pairwise), reducing complexity from exponential `O(2^N)` to polynomial `O(N^k)`.
-    * **Parallel Execution**: When compiled with the `parallel` feature flag, the main decomposition loop runs in
-      parallel across all available CPU cores using `rayon`.
+    * **Parallel Execution**: When compiled with the `parallel` feature flag, the main decomposition loop of the SURD algorithm and the feature selection loops of the mRMR algorithm run in parallel across all available CPU cores using `rayon`.
 
 ## Installation
 

deep_causality_algorithms/README.md

@@ -4,7 +4,7 @@
  */
 
 use criterion::{Criterion, criterion_group, criterion_main};
-use deep_causality_algorithms::mrmr::mrmr_features_selector;
+use deep_causality_algorithms::mrmr::{mrmr_features_selector, mrmr_features_selector_cdl};
 use deep_causality_tensor::CausalTensor;
 
 fn generate_test_tensor(rows: usize, cols: usize) -> CausalTensor<f64> {
@@ -15,21 +15,40 @@ fn generate_test_tensor(rows: usize, cols: usize) -> CausalTensor<f64> {
     CausalTensor::new(data, vec![rows, cols]).unwrap()
 }
 
+fn generate_test_tensor_cdl(rows: usize, cols: usize) -> CausalTensor<Option<f64>> {
+    let mut data = Vec::with_capacity(rows * cols);
+    for i in 0..(rows * cols) {
+        if (i * 3 + i / 5) % 13 < 2 {
+            // A more complex pattern
+            data.push(None);
+        } else {
+            data.push(Some(i as f64));
+        }
+    }
+    CausalTensor::new(data, vec![rows, cols]).unwrap()
+}
+
 fn mrmr_benchmark(c: &mut Criterion) {
     let mut group = c.benchmark_group("mRMR Feature Selector");
 
-    let rows = 100;
-    let cols = 20;
-    let num_features_to_select = 5;
+    let rows = 1000;
+    let cols = 100;
+    let num_features_to_select = 10;
     let target_col = cols - 1;
 
-    // Benchmark the new implementation
-    group.bench_function("mrmr_features_selector_new_impl", |b| {
-        let tensor = generate_test_tensor(rows, cols);
+    // Benchmark the standard implementation
+    group.bench_function("mrmr_features_selector", |b| {
+        let mut tensor = generate_test_tensor(rows, cols);
+        b.iter(|| {
+            mrmr_features_selector(&mut tensor, num_features_to_select, target_col).unwrap();
+        });
+    });
+
+    // Benchmark the cdl implementation
+    group.bench_function("mrmr_features_selector_cdl", |b| {
+        let tensor = generate_test_tensor_cdl(rows, cols);
         b.iter(|| {
-            // Clone the tensor for each iteration to ensure a fresh state
-            let mut cloned_tensor = tensor.clone();
-            mrmr_features_selector(&mut cloned_tensor, num_features_to_select, target_col).unwrap();
+            mrmr_features_selector_cdl(&tensor, num_features_to_select, target_col).unwrap();
         });
     });