sklears-utils 0.1.1

//! Property-based tests for utility functions
//!
//! These tests verify mathematical properties and invariants that should hold
//! for all inputs using proptest.

use crate::array_utils::{label_counts, unique_labels};
use crate::data_generation::*;
use crate::multiclass::type_of_target;
use crate::validation::check_finite;
use proptest::prelude::*;
use scirs2_core::ndarray::{Array1, Array2};

proptest! {
    #[test]
    fn test_make_classification_properties(
        n_samples in 10..100usize,
        n_features in 2..10usize,
        n_classes in 2..5usize,
        flip_y in 0.0..0.3f64,
        class_sep in 0.5..3.0f64
    ) {
        let result = make_classification(
            n_samples, n_features, n_classes,
            None, None, flip_y, class_sep, Some(42)
        );

        if let Ok((x, y)) = result {
            // Check shapes
            prop_assert_eq!(x.shape(), &[n_samples, n_features]);
            prop_assert_eq!(y.len(), n_samples);

            // Check that all labels are valid
            for &label in y.iter() {
                prop_assert!(label >= 0 && label < n_classes as i32);
            }

            // Check that at least some classes are present (unless heavily flipped)
            let unique_classes: std::collections::HashSet<i32> = y.iter().copied().collect();
            if flip_y < 0.9 {
                prop_assert!(!unique_classes.is_empty());
            }

            // All feature values should be finite
            for &val in x.iter() {
                prop_assert!(val.is_finite());
            }
        }
    }

    #[test]
    fn test_make_regression_properties(
        n_samples in 10..100usize,
        n_features in 2..10usize,
        noise in 0.0..2.0f64,
        bias in -10.0..10.0f64
    ) {
        let result = make_regression(
            n_samples, n_features, Some(n_features),
            noise, bias, Some(42)
        );

        if let Ok((x, y)) = result {
            // Check shapes
            prop_assert_eq!(x.shape(), &[n_samples, n_features]);
            prop_assert_eq!(y.len(), n_samples);

            // All values should be finite
            for &val in x.iter() {
                prop_assert!(val.is_finite());
            }
            for &val in y.iter() {
                prop_assert!(val.is_finite());
            }
        }
    }

    #[test]
    fn test_make_blobs_properties(
        n_samples in 10..100usize,
        n_features in 2..8usize,
        centers in 2..6usize,
        cluster_std in 0.1..3.0f64
    ) {
        let result = make_blobs(
            n_samples, n_features, Some(centers),
            cluster_std, (-10.0, 10.0), Some(42)
        );

        if let Ok((x, y)) = result {
            // Check shapes
            prop_assert_eq!(x.shape(), &[n_samples, n_features]);
            prop_assert_eq!(y.len(), n_samples);

            // Check that all cluster labels are valid
            for &label in y.iter() {
                prop_assert!(label >= 0 && label < centers as i32);
            }

            // All cluster IDs should be present
            let unique_clusters: std::collections::HashSet<i32> = y.iter().copied().collect();
            prop_assert_eq!(unique_clusters.len(), centers);

            // All values should be finite
            for &val in x.iter() {
                prop_assert!(val.is_finite());
            }
        }
    }

    // Removed check_range test - function signature is different than expected

    #[test]
    fn test_check_finite_2d_properties(
        values in prop::collection::vec(-100.0..100.0f64, 4..40)
    ) {
        let size = values.len();
        let rows = (size as f64).sqrt() as usize;
        let cols = size / rows;
        let actual_size = rows * cols;

        if actual_size > 0 {
            let trimmed_values: Vec<f64> = values.into_iter().take(actual_size).collect();
            let array = Array2::from_shape_vec((rows, cols), trimmed_values).expect("operation should succeed");

            // All finite values should pass
            prop_assert!(check_finite(&array).is_ok());
        }
    }

    // Removed test_check_finite_properties since check_finite only works with 2D arrays

    #[test]
    fn test_unique_labels_properties(
        labels in prop::collection::vec(0..10i32, 1..20)
    ) {
        let array = Array1::from_vec(labels.clone());
        let unique = unique_labels(&array);

        // Unique should contain all distinct values
        let expected_unique: std::collections::HashSet<i32> = labels.into_iter().collect();
        let actual_unique: std::collections::HashSet<i32> = unique.into_iter().collect();

        prop_assert_eq!(expected_unique, actual_unique);
    }

    #[test]
    fn test_label_counts_properties(
        labels in prop::collection::vec(0..5i32, 1..50)
    ) {
        let array = Array1::from_vec(labels.clone());
        let counts = label_counts(&array);

        // Sum of counts should equal total number of labels
        let total_count: usize = counts.values().sum();
        prop_assert_eq!(total_count, labels.len());

        // Each unique label should have a positive count
        for &count in counts.values() {
            prop_assert!(count > 0);
        }

        // Manual count verification for a few labels
        let manual_counts: std::collections::HashMap<i32, usize> =
            labels.iter().fold(std::collections::HashMap::new(), |mut acc, &label| {
                *acc.entry(label).or_insert(0) += 1;
                acc
            });

        prop_assert_eq!(counts, manual_counts);
    }

    #[test]
    fn test_type_of_target_properties(
        target_type in 0..3usize,
        size in 5..30usize
    ) {
        let y = match target_type {
            0 => Array1::from_vec((0..size).map(|i| (i % 2) as i32).collect()), // binary
            1 => Array1::from_vec((0..size).map(|i| (i % 5) as i32).collect()), // multiclass
            _ => Array1::from_vec(vec![0i32; size]), // constant
        };

        let target_type_result = type_of_target(&y);

        // Should return a valid target type string
        if let Ok(target_type_str) = target_type_result {
            prop_assert!(target_type_str == "binary" ||
                        target_type_str == "multiclass" ||
                        target_type_str == "unknown");
        }
    }

    #[test]
    fn test_train_test_split_properties(
        n_samples in 10..100usize,
        n_features in 2..10usize,
        test_size in 0.1..0.5f64
    ) {
        let _x = Array2::<f64>::zeros((n_samples, n_features));
        let _y = Array1::from_vec((0..n_samples).map(|i| (i % 3) as i32).collect());

        use crate::random::train_test_split_indices;
        let result = train_test_split_indices(n_samples, test_size, true, Some(42));

        if let Ok((train_indices, test_indices)) = result {
            // Check that all samples are accounted for
            let total_samples = train_indices.len() + test_indices.len();
            prop_assert_eq!(total_samples, n_samples);

            // Check that test size is approximately correct
            let actual_test_ratio = test_indices.len() as f64 / n_samples as f64;
            let tolerance = 0.1; // Allow 10% tolerance
            prop_assert!((actual_test_ratio - test_size).abs() <= tolerance);

            // Check that indices are unique and valid
            let mut all_indices = train_indices.clone();
            all_indices.extend(test_indices.clone());
            all_indices.sort_unstable();
            let expected_indices: Vec<usize> = (0..n_samples).collect();
            prop_assert_eq!(all_indices, expected_indices);
        }
    }

    #[test]
    fn test_bootstrap_indices_properties(
        n_samples in 10..100usize
    ) {
        use crate::random::bootstrap_indices;
        let indices = bootstrap_indices(n_samples, Some(42));

        // Check length
        prop_assert_eq!(indices.len(), n_samples);

        // Check that all indices are valid
        for &idx in indices.iter() {
            prop_assert!(idx < n_samples);
        }

        // Bootstrap should allow repetition, so with enough samples,
        // we should see some repeated indices
        if n_samples > 10 {
            let unique_indices: std::collections::HashSet<usize> = indices.into_iter().collect();
            // Usually we should see some repetition in bootstrap samples
            prop_assert!(unique_indices.len() <= n_samples);
        }
    }

    #[test]
    fn test_stratified_split_properties(
        n_samples in 20..100usize,
        n_classes in 2..5usize,
        test_size in 0.2..0.5f64
    ) {
        use crate::random::stratified_split_indices;

        // Create stratified labels
        let labels: Vec<i32> = (0..n_samples)
            .map(|i| (i % n_classes) as i32)
            .collect();

        let result = stratified_split_indices(&labels, test_size, Some(42));

        if let Ok((train_indices, test_indices)) = result {
            // Check total samples
            prop_assert_eq!(train_indices.len() + test_indices.len(), n_samples);

            // Check that stratification is maintained
            let train_labels: Vec<i32> = train_indices.iter()
                .map(|&i| labels[i])
                .collect();
            let test_labels: Vec<i32> = test_indices.iter()
                .map(|&i| labels[i])
                .collect();

            // Each class should be represented in both sets (if possible)
            let train_classes: std::collections::HashSet<i32> = train_labels.into_iter().collect();
            let test_classes: std::collections::HashSet<i32> = test_labels.into_iter().collect();

            // At least some overlap in classes between train and test
            let overlap: std::collections::HashSet<_> = train_classes.intersection(&test_classes).collect();
            if test_size < 0.9 && n_samples >= n_classes * 4 {
                prop_assert!(!overlap.is_empty());
            }
        }
    }

    #[test]
    fn test_shuffle_indices_properties(
        n_samples in 5..50usize
    ) {
        use crate::random::random_permutation;

        let indices = random_permutation(n_samples, Some(42));

        // Check length
        prop_assert_eq!(indices.len(), n_samples);

        // Check that all original indices are present
        let mut sorted_indices = indices.clone();
        sorted_indices.sort_unstable();
        let expected: Vec<usize> = (0..n_samples).collect();
        prop_assert_eq!(sorted_indices, expected);

        // Check uniqueness (no duplicates)
        let unique_indices: std::collections::HashSet<usize> = indices.into_iter().collect();
        prop_assert_eq!(unique_indices.len(), n_samples);
    }

    #[test]
    fn test_k_fold_indices_properties(
        n_samples in 10..100usize,
        n_splits in 2..10usize
    ) {
        use crate::random::k_fold_indices;

        let result = k_fold_indices(n_samples, n_splits, true, Some(42));

        if let Ok(folds) = result {
            prop_assert_eq!(folds.len(), n_splits);

            // Check that all samples are used exactly once across all folds
            let mut all_test_indices = Vec::new();
            for (_, test_indices) in &folds {
                all_test_indices.extend(test_indices.iter().copied());
            }

            all_test_indices.sort_unstable();
            let expected: Vec<usize> = (0..n_samples).collect();
            prop_assert_eq!(all_test_indices, expected);

            // Check that fold sizes are balanced
            let mut fold_sizes: Vec<usize> = folds.iter().map(|(_, test)| test.len()).collect();
            fold_sizes.sort_unstable();

            // Max difference between fold sizes should be at most 1
            if fold_sizes.len() > 1 {
                prop_assert!(fold_sizes[fold_sizes.len() - 1] - fold_sizes[0] <= 1);
            }
        }
    }

    #[test]
    fn test_cross_validation_split_consistency(
        n_samples in 20..80usize,
        n_splits in 3..8usize
    ) {
        use crate::random::k_fold_indices;

        // Test that K-fold provides consistent splits
        let result1 = k_fold_indices(n_samples, n_splits, true, Some(42));
        let result2 = k_fold_indices(n_samples, n_splits, true, Some(42));

        if let (Ok(folds1), Ok(folds2)) = (result1, result2) {
            // Same seed should produce identical results
            prop_assert_eq!(folds1.len(), folds2.len());

            for (fold1, fold2) in folds1.iter().zip(folds2.iter()) {
                prop_assert_eq!(&fold1.0, &fold2.0); // train indices
                prop_assert_eq!(&fold1.1, &fold2.1); // test indices
            }
        }
    }

    #[test]
    fn test_array_validation_properties(
        n_features in 2..10usize,
        n_samples in 5..20usize
    ) {
        use crate::validation::{check_array_2d, check_consistent_length_xy};

        // Create valid array
        let x = Array2::<f64>::zeros((n_samples, n_features));
        let y = Array1::<i32>::zeros(n_samples);

        // Valid arrays should pass validation
        prop_assert!(check_array_2d(&x).is_ok());
        prop_assert!(check_consistent_length_xy(&x, &y).is_ok());

        // Check inconsistent length detection
        let y_wrong = Array1::<i32>::zeros(n_samples + 1);
        prop_assert!(check_consistent_length_xy(&x, &y_wrong).is_err());
    }

    #[test]
    fn test_distance_computation_properties(
        dim in 2..8usize
    ) {
        use crate::metrics::{euclidean_distance, manhattan_distance};

        let point1 = Array1::<f64>::zeros(dim);
        let point2 = Array1::<f64>::ones(dim);

        // Distance should be non-negative
        let euclidean_dist = euclidean_distance(&point1, &point2);
        let manhattan_dist = manhattan_distance(&point1, &point2);

        prop_assert!(euclidean_dist >= 0.0);
        prop_assert!(manhattan_dist >= 0.0);

        // Distance from point to itself should be zero
        let self_euclidean = euclidean_distance(&point1, &point1);
        let self_manhattan = manhattan_distance(&point1, &point1);

        prop_assert!((self_euclidean - 0.0).abs() < 1e-10);
        prop_assert!((self_manhattan - 0.0).abs() < 1e-10);

        // Manhattan distance should equal sqrt(dim) for unit vectors
        prop_assert!((manhattan_dist - dim as f64).abs() < 1e-10);
    }

    #[test]
    fn test_data_scaling_properties(
        n_samples in 10..50usize,
        n_features in 2..10usize,
        scale in 0.1..10.0f64
    ) {
        // Create test data with known scale
        let mut x = Array2::<f64>::zeros((n_samples, n_features));
        for i in 0..n_samples {
            for j in 0..n_features {
                x[[i, j]] = scale * (i as f64 + j as f64);
            }
        }

        // Check that data has expected properties
        prop_assert!(x.iter().all(|&val| val.is_finite()));

        // Check that scaling preserves relationships
        let max_val = x.iter().fold(f64::NEG_INFINITY, |a, &b| a.max(b));
        let min_val = x.iter().fold(f64::INFINITY, |a, &b| a.min(b));

        prop_assert!(max_val >= min_val);
        if scale > 0.0 {
            prop_assert!(max_val > min_val || n_samples <= 1 || n_features <= 1);
        }
    }

}

/// Unit test for environment detection properties (not property-based)
#[allow(non_snake_case)]
#[cfg(test)]
#[test]
fn test_environment_detection_basic() {
    use crate::environment::{EnvironmentInfo, FeatureChecker, HardwareDetector};

    let hw = HardwareDetector::new();
    let checker = FeatureChecker::new();
    let env = EnvironmentInfo::detect();

    // Basic hardware detection properties
    assert!(hw.cpu_cores() >= 1);
    assert!(hw.logical_cores() >= hw.cpu_cores());
    assert!(hw.total_memory() > 0);
    assert!(!hw.cpu_architecture().is_empty());

    // Feature checking properties
    assert!(checker.recommended_thread_count() >= 1);
    assert!(checker.recommended_simd_width() >= 1);
    assert!(checker.high_resolution_timer_available());

    // Environment summary should be non-empty
    let summary = env.summary();
    assert!(!summary.is_empty());
    assert!(summary.contains("Environment Summary"));
}

/// Comprehensive testing framework utilities
pub mod testing_framework {
    use super::*;
    use std::time::Duration;

    /// Test suite runner for comprehensive validation
    pub struct TestSuite {
        pub name: String,
        pub tests: Vec<Box<dyn Fn() -> Result<(), String> + Send + Sync>>,
        pub setup: Option<Box<dyn Fn() -> Result<(), String> + Send + Sync>>,
        pub teardown: Option<Box<dyn Fn() -> Result<(), String> + Send + Sync>>,
        #[allow(dead_code)] // intentionally deferred: test timeout enforcement pending
        pub timeout: Duration,
    }

    impl TestSuite {
        /// Create a new test suite
        pub fn new(name: &str) -> Self {
            Self {
                name: name.to_string(),
                tests: Vec::new(),
                setup: None,
                teardown: None,
                timeout: Duration::from_secs(60),
            }
        }

        /// Add a test to the suite
        pub fn add_test<F>(&mut self, test: F)
        where
            F: Fn() -> Result<(), String> + Send + Sync + 'static,
        {
            self.tests.push(Box::new(test));
        }

        /// Set setup function
        #[allow(dead_code)] // intentionally deferred: suite lifecycle hooks pending
        pub fn set_setup<F>(&mut self, setup: F)
        where
            F: Fn() -> Result<(), String> + Send + Sync + 'static,
        {
            self.setup = Some(Box::new(setup));
        }

        /// Set teardown function
        #[allow(dead_code)] // intentionally deferred: suite lifecycle hooks pending
        pub fn set_teardown<F>(&mut self, teardown: F)
        where
            F: Fn() -> Result<(), String> + Send + Sync + 'static,
        {
            self.teardown = Some(Box::new(teardown));
        }

        /// Run all tests in the suite
        pub fn run(&self) -> TestResults {
            let mut results = TestResults::new(&self.name);

            // Run setup
            if let Some(ref setup) = self.setup {
                if let Err(e) = setup() {
                    results.add_failure(format!("Setup failed: {e}"));
                    return results;
                }
            }

            // Run tests
            for (i, test) in self.tests.iter().enumerate() {
                let test_name = format!("test_{i}");
                match test() {
                    Ok(()) => results.add_success(test_name),
                    Err(e) => results.add_failure(format!("{test_name}: {e}")),
                }
            }

            // Run teardown
            if let Some(ref teardown) = self.teardown {
                if let Err(e) = teardown() {
                    results.add_failure(format!("Teardown failed: {e}"));
                }
            }

            results
        }
    }

    /// Test results container
    #[derive(Debug, Clone)]
    pub struct TestResults {
        #[allow(dead_code)] // intentionally deferred: suite name readout pending
        pub suite_name: String,
        pub successes: Vec<String>,
        pub failures: Vec<String>,
        #[allow(dead_code)] // intentionally deferred: elapsed time tracking pending
        pub duration: Duration,
    }

    impl TestResults {
        pub fn new(suite_name: &str) -> Self {
            Self {
                suite_name: suite_name.to_string(),
                successes: Vec::new(),
                failures: Vec::new(),
                duration: Duration::new(0, 0),
            }
        }

        pub fn add_success(&mut self, test_name: String) {
            self.successes.push(test_name);
        }

        pub fn add_failure(&mut self, error: String) {
            self.failures.push(error);
        }

        pub fn total_tests(&self) -> usize {
            self.successes.len() + self.failures.len()
        }

        #[allow(dead_code)] // intentionally deferred: test result analysis pending
        pub fn success_rate(&self) -> f64 {
            if self.total_tests() == 0 {
                0.0
            } else {
                self.successes.len() as f64 / self.total_tests() as f64
            }
        }

        pub fn is_successful(&self) -> bool {
            self.failures.is_empty()
        }

        #[allow(dead_code)] // intentionally deferred: test result summary rendering pending
        pub fn summary(&self) -> String {
            format!(
                "Test Suite: {}\n\
                Total Tests: {}\n\
                Successes: {}\n\
                Failures: {}\n\
                Success Rate: {:.1}%\n\
                Duration: {:?}",
                self.suite_name,
                self.total_tests(),
                self.successes.len(),
                self.failures.len(),
                self.success_rate() * 100.0,
                self.duration
            )
        }
    }

    /// Stress testing utilities
    pub struct StressTester {
        pub iterations: usize,
        pub concurrent_threads: usize,
        #[allow(dead_code)] // intentionally deferred: stress test timeout enforcement pending
        pub timeout: Duration,
    }

    impl StressTester {
        pub fn new() -> Self {
            Self {
                iterations: 1000,
                concurrent_threads: num_cpus::get(),
                timeout: Duration::from_secs(300),
            }
        }

        pub fn with_iterations(mut self, iterations: usize) -> Self {
            self.iterations = iterations;
            self
        }

        pub fn with_threads(mut self, threads: usize) -> Self {
            self.concurrent_threads = threads;
            self
        }

        /// Run stress test on a function
        pub fn stress_test<F, T>(&self, test_fn: F) -> Result<Vec<T>, String>
        where
            F: Fn() -> Result<T, String> + Send + Sync + Clone + 'static,
            T: Send + Clone + 'static,
        {
            use std::sync::{Arc, Mutex};
            use std::thread;

            let results = Arc::new(Mutex::new(Vec::<T>::new()));
            let iterations_per_thread = self.iterations / self.concurrent_threads;
            let test_fn = Arc::new(test_fn);

            let handles: Vec<_> = (0..self.concurrent_threads)
                .map(|_| {
                    let test_fn = Arc::clone(&test_fn);
                    let results = Arc::clone(&results);

                    thread::spawn(move || {
                        for _ in 0..iterations_per_thread {
                            match test_fn() {
                                Ok(result) => {
                                    if let Ok(mut results) = results.lock() {
                                        results.push(result);
                                    }
                                }
                                Err(e) => return Err(e),
                            }
                        }
                        Ok(())
                    })
                })
                .collect();

            // Wait for all threads to complete
            for handle in handles {
                handle
                    .join()
                    .map_err(|_| "Thread panicked".to_string())?
                    .map_err(|e| format!("Stress test failed: {e}"))?;
            }

            let results = results
                .lock()
                .map_err(|_| "Failed to acquire lock".to_string())?;
            Ok(results.to_vec())
        }
    }

    /// Property-based test generator
    #[allow(dead_code)] // intentionally deferred: generator strategies not yet used in proptest harness
    pub struct PropertyTestGenerator;

    impl PropertyTestGenerator {
        /// Generate test data for array operations
        #[allow(dead_code)] // intentionally deferred: strategy not yet wired to proptest harness
        pub fn array_data_strategy() -> impl Strategy<Value = (usize, usize)> {
            (10..100usize, 2..20usize)
        }

        /// Generate test data for optimization algorithms
        #[allow(dead_code)] // intentionally deferred: strategy not yet wired to proptest harness
        pub fn optimization_strategy() -> impl Strategy<Value = (f64, f64, usize)> {
            (1e-8..1e-2f64, 1e-8..1e-2f64, 10..1000usize)
        }

        /// Generate test data for spatial operations
        #[allow(dead_code)] // intentionally deferred: strategy not yet wired to proptest harness
        pub fn spatial_strategy() -> impl Strategy<Value = (usize, usize, f64)> {
            (2..4usize, 10..100usize, 1.0..1000.0f64)
        }

        /// Generate test data for statistical tests
        #[allow(dead_code)] // intentionally deferred: strategy not yet wired to proptest harness
        pub fn statistical_strategy() -> impl Strategy<Value = (usize, f64, f64)> {
            (10..1000usize, -100.0..100.0f64, 0.1..10.0f64)
        }
    }

    #[allow(non_snake_case)]
    #[cfg(test)]
    mod framework_tests {
        use super::*;

        #[test]
        fn test_test_suite() {
            let mut suite = TestSuite::new("example_suite");

            suite.add_test(|| Ok(()));
            suite.add_test(|| Err("Test failure".to_string()));
            suite.add_test(|| Ok(()));

            let results = suite.run();

            assert_eq!(results.total_tests(), 3);
            assert_eq!(results.successes.len(), 2);
            assert_eq!(results.failures.len(), 1);
            assert!(!results.is_successful());
        }

        #[test]
        fn test_stress_tester() {
            let tester = StressTester::new().with_iterations(100).with_threads(2);

            let results = tester.stress_test(|| Ok(42));
            assert!(results.is_ok());

            let values = results.expect("operation should succeed");
            assert_eq!(values.len(), 100);
            assert!(values.iter().all(|&x| x == 42));
        }
    }
}

#[allow(non_snake_case)]
#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_property_test_setup() {
        // Simple test to ensure property test framework is working
        let (x, y) = make_classification(20, 3, 2, None, None, 0.0, 1.0, Some(42))
            .expect("operation should succeed");

        assert_eq!(x.shape(), &[20, 3]);
        assert_eq!(y.len(), 20);
    }
}