aprender-core 0.29.1

proptest! {
    #![proptest_config(ProptestConfig::with_cases(100))]


    // Vector properties
    #[test]
    fn vector_sum_is_associative(a in vector_strategy(10), b in vector_strategy(10)) {
        let sum_ab = (&a + &b).sum();
        let sum_ba = (&b + &a).sum();
        prop_assert!((sum_ab - sum_ba).abs() < 1e-4);
    }

    #[test]
    fn vector_dot_is_commutative(a in vector_strategy(10), b in vector_strategy(10)) {
        let dot_ab = a.dot(&b);
        let dot_ba = b.dot(&a);
        prop_assert!((dot_ab - dot_ba).abs() < 1e-4);
    }

    #[test]
    fn vector_norm_is_non_negative(v in vector_strategy(10)) {
        prop_assert!(v.norm() >= 0.0);
    }

    #[test]
    fn vector_scalar_mul_distributes(v in vector_strategy(10), s in -10.0f32..10.0) {
        let scaled = v.mul_scalar(s);
        let expected_sum = v.sum() * s;
        prop_assert!((scaled.sum() - expected_sum).abs() < 1e-3);
    }

    #[test]
    fn vector_elementwise_mul_is_commutative(a in vector_strategy(10), b in vector_strategy(10)) {
        let mul_ab = &a * &b;
        let mul_ba = &b * &a;
        for i in 0..10 {
            prop_assert!((mul_ab[i] - mul_ba[i]).abs() < 1e-4);
        }
    }

    #[test]
    fn vector_elementwise_mul_with_ones_is_identity(v in vector_strategy(10)) {
        let ones = Vector::<f32>::ones(10);
        let result = &v * &ones;
        for i in 0..10 {
            prop_assert!((result[i] - v[i]).abs() < 1e-6);
        }
    }

    #[test]
    fn vector_elementwise_mul_with_zeros_is_zero(v in vector_strategy(10)) {
        let zeros = Vector::<f32>::zeros(10);
        let result = &v * &zeros;
        for i in 0..10 {
            prop_assert!((result[i]).abs() < 1e-6);
        }
    }

    // Matrix properties
    #[test]
    fn matrix_transpose_involution(m in matrix_strategy(5, 5)) {
        let m_t = m.transpose();
        let m_tt = m_t.transpose();
        for i in 0..5 {
            for j in 0..5 {
                prop_assert!((m.get(i, j) - m_tt.get(i, j)).abs() < 1e-6);
            }
        }
    }

    #[test]
    fn matrix_shape_preserved_by_add(a in matrix_strategy(4, 3), b in matrix_strategy(4, 3)) {
        let c = a.add(&b).expect("Test data should be valid");
        prop_assert_eq!(c.shape(), (4, 3));
    }

    #[test]
    fn matrix_matmul_shape(a in matrix_strategy(3, 4), b in matrix_strategy(4, 2)) {
        let c = a.matmul(&b).expect("Test data should be valid");
        prop_assert_eq!(c.shape(), (3, 2));
    }

    // DataFrame properties
    #[test]
    fn dataframe_select_preserves_rows(
        data in proptest::collection::vec(-100.0f32..100.0, 15)
    ) {
        let columns = vec![
            ("a".to_string(), Vector::from_slice(&data[0..5])),
            ("b".to_string(), Vector::from_slice(&data[5..10])),
            ("c".to_string(), Vector::from_slice(&data[10..15])),
        ];
        let df = DataFrame::new(columns).expect("Test data should be valid");
        let selected = df.select(&["a", "c"]).expect("Test data should be valid");
        prop_assert_eq!(selected.n_rows(), df.n_rows());
    }

    #[test]
    fn dataframe_to_matrix_shape(
        data in proptest::collection::vec(-100.0f32..100.0, 15)
    ) {
        let columns = vec![
            ("a".to_string(), Vector::from_slice(&data[0..5])),
            ("b".to_string(), Vector::from_slice(&data[5..10])),
            ("c".to_string(), Vector::from_slice(&data[10..15])),
        ];
        let df = DataFrame::new(columns).expect("Test data should be valid");
        let matrix = df.to_matrix();
        prop_assert_eq!(matrix.shape(), (5, 3));
    }

    // Metrics properties
    #[test]
    fn r_squared_perfect_prediction(y in vector_strategy(10)) {
        let r2 = r_squared(&y, &y);
        // Perfect prediction should give R² = 1 (or very close)
        prop_assert!((r2 - 1.0).abs() < 1e-6 || y.variance() == 0.0);
    }

    #[test]
    fn mse_is_non_negative(y_true in vector_strategy(10), y_pred in vector_strategy(10)) {
        let error = mse(&y_pred, &y_true);
        prop_assert!(error >= 0.0);
    }

    #[test]
    fn mae_is_non_negative(y_true in vector_strategy(10), y_pred in vector_strategy(10)) {
        let error = mae(&y_pred, &y_true);
        prop_assert!(error >= 0.0);
    }

    #[test]
    fn rmse_equals_sqrt_mse(y_true in vector_strategy(10), y_pred in vector_strategy(10)) {
        let mse_val = mse(&y_pred, &y_true);
        let rmse_val = rmse(&y_pred, &y_true);
        prop_assert!((rmse_val - mse_val.sqrt()).abs() < 1e-6);
    }

    // Linear regression properties
    #[test]
    fn linear_regression_coefficients_length(n_features in 1usize..5) {
        // Create well-conditioned data
        let n_samples = n_features + 3;
        let mut x_data = vec![0.0; n_samples * n_features];
        let mut y_data = vec![0.0; n_samples];

        // Create identity-like pattern to ensure positive definiteness
        for i in 0..n_samples {
            for j in 0..n_features {
                x_data[i * n_features + j] = if i == j { 1.0 } else { 0.1 * (i + j) as f32 };
            }
            y_data[i] = (i + 1) as f32;
        }

        let x = Matrix::from_vec(n_samples, n_features, x_data).expect("Test data should be valid");
        let y = Vector::from_vec(y_data);

        let mut model = LinearRegression::new();
        if model.fit(&x, &y).is_ok() {
            prop_assert_eq!(model.coefficients().len(), n_features);
        }
    }

    // K-Means properties
    #[test]
    fn kmeans_labels_valid(n_clusters in 1usize..4) {
        // Create data with clear clusters
        let n_samples = n_clusters * 5;
        let mut data = Vec::with_capacity(n_samples * 2);

        for k in 0..n_clusters {
            for i in 0..5 {
                data.push((k * 10) as f32 + i as f32 * 0.1);
                data.push((k * 10) as f32 + i as f32 * 0.1);
            }
        }

        let matrix = Matrix::from_vec(n_samples, 2, data).expect("Test data should be valid");
        let mut kmeans = KMeans::new(n_clusters).with_random_state(42);
        kmeans.fit(&matrix).expect("Test data should be valid");

        let labels = kmeans.predict(&matrix);

        // All labels should be valid cluster indices
        for &label in &labels {
            prop_assert!(label < n_clusters);
        }
    }

    #[test]
    fn kmeans_inertia_non_negative(n_clusters in 1usize..3) {
        let n_samples = n_clusters * 3;
        let data: Vec<f32> = (0..n_samples * 2).map(|i| i as f32).collect();
        let matrix = Matrix::from_vec(n_samples, 2, data).expect("Test data should be valid");

        let mut kmeans = KMeans::new(n_clusters).with_random_state(42);
        kmeans.fit(&matrix).expect("Test data should be valid");

        prop_assert!(kmeans.inertia() >= 0.0);
    }

    // Train-test split properties
    #[test]
    fn train_test_split_preserves_samples(n_samples in 10usize..50) {
        let x_data: Vec<f32> = (0..n_samples * 2).map(|i| i as f32).collect();
        let y_data: Vec<f32> = (0..n_samples).map(|i| i as f32).collect();

        let x = Matrix::from_vec(n_samples, 2, x_data).expect("Test data should be valid");
        let y = Vector::from_vec(y_data);

        let (x_train, x_test, y_train, y_test) = train_test_split(&x, &y, 0.2, Some(42)).expect("Test data should be valid");

        // Total samples should be preserved
        let total = x_train.n_rows() + x_test.n_rows();
        prop_assert_eq!(total, n_samples);
        prop_assert_eq!(y_train.len() + y_test.len(), n_samples);
    }

    #[test]
    fn train_test_split_ratio_approximate(test_size in 0.1f32..0.5) {
        let n_samples = 100;
        let x_data: Vec<f32> = (0..n_samples * 2).map(|i| i as f32).collect();
        let y_data: Vec<f32> = (0..n_samples).map(|i| i as f32).collect();

        let x = Matrix::from_vec(n_samples, 2, x_data).expect("Test data should be valid");
        let y = Vector::from_vec(y_data);

        let (_, x_test, _, _) = train_test_split(&x, &y, test_size, Some(42)).expect("Test data should be valid");

        // Test set should be approximately the right size
        let actual_ratio = x_test.n_rows() as f32 / n_samples as f32;
        prop_assert!((actual_ratio - test_size).abs() < 0.1);
    }

    #[test]
    fn kfold_splits_cover_all_samples(k in 2usize..6) {
        let n_samples = 20;
        let kfold = KFold::new(k);

        let mut seen = vec![false; n_samples];
        for (_, test_indices) in kfold.split(n_samples) {
            for &idx in &test_indices {
                seen[idx] = true;
            }
        }

        // Every sample should appear in exactly one test fold
        for (i, &s) in seen.iter().enumerate() {
            prop_assert!(s, "Sample {} not in any test fold", i);
        }
    }

    // Decision tree properties
    #[test]
    fn decision_tree_predictions_valid(n_classes in 2usize..4) {
        // Create well-separated data
        let n_per_class = 5;
        let n_samples = n_classes * n_per_class;
        let mut data = Vec::with_capacity(n_samples * 2);
        let mut labels = Vec::with_capacity(n_samples);

        for class in 0..n_classes {
            for i in 0..n_per_class {
                data.push((class * 10) as f32 + i as f32 * 0.1);
                data.push((class * 10) as f32 + i as f32 * 0.1);
                labels.push(class);
            }
        }

        let x = Matrix::from_vec(n_samples, 2, data).expect("Test data should be valid");
        let mut tree = DecisionTreeClassifier::new().with_max_depth(5);
        tree.fit(&x, &labels).expect("Test data should be valid");

        let predictions = tree.predict(&x);

        // All predictions should be valid class indices
        for &pred in &predictions {
            prop_assert!(pred < n_classes);
        }
    }

    // StandardScaler properties
    #[test]
    fn standard_scaler_produces_zero_mean(data in matrix_strategy(10, 3)) {
        let mut scaler = StandardScaler::new();
        let transformed = scaler.fit_transform(&data).expect("Test data should be valid");

        let (n_rows, n_cols) = transformed.shape();
        for j in 0..n_cols {
            let mut sum = 0.0;
            for i in 0..n_rows {
                sum += transformed.get(i, j);
            }
            let mean = sum / n_rows as f32;
            prop_assert!(mean.abs() < 1e-4, "Column {} mean should be ~0, got {}", j, mean);
        }
    }

    #[test]
    fn standard_scaler_produces_unit_variance(data in matrix_strategy(10, 3)) {
        let mut scaler = StandardScaler::new();
        let transformed = scaler.fit_transform(&data).expect("Test data should be valid");

        let (n_rows, n_cols) = transformed.shape();
        for j in 0..n_cols {
            // Compute variance
            let mut sum = 0.0;
            for i in 0..n_rows {
                sum += transformed.get(i, j);
            }
            let mean = sum / n_rows as f32;

            let mut var_sum = 0.0;
            for i in 0..n_rows {
                let diff = transformed.get(i, j) - mean;
                var_sum += diff * diff;
            }
            let std = (var_sum / n_rows as f32).sqrt();

            // std should be 1 (or 0 if constant column)
            prop_assert!(std < 1e-4 || (std - 1.0).abs() < 1e-4,
                "Column {} std should be ~0 or ~1, got {}", j, std);
        }
    }

    #[test]
    fn standard_scaler_inverse_recovers_data(data in matrix_strategy(8, 2)) {
        let mut scaler = StandardScaler::new();
        let transformed = scaler.fit_transform(&data).expect("Test data should be valid");
        let recovered = scaler.inverse_transform(&transformed).expect("Test data should be valid");

        let (n_rows, n_cols) = data.shape();
        for i in 0..n_rows {
            for j in 0..n_cols {
                prop_assert!(
                    (data.get(i, j) - recovered.get(i, j)).abs() < 1e-3,
                    "Mismatch at ({}, {}): expected {}, got {}",
                    i, j, data.get(i, j), recovered.get(i, j)
                );
            }
        }
    }

    // MinMaxScaler properties
    #[test]
    fn minmax_scaler_bounds_to_range(data in matrix_strategy(10, 3)) {
        let mut scaler = MinMaxScaler::new();
        let transformed = scaler.fit_transform(&data).expect("Test data should be valid");

        let (n_rows, n_cols) = transformed.shape();
        for j in 0..n_cols {
            let mut min_val = f32::INFINITY;
            let mut max_val = f32::NEG_INFINITY;
            for i in 0..n_rows {
                let val = transformed.get(i, j);
                if val < min_val {
                    min_val = val;
                }
                if val > max_val {
                    max_val = val;
                }
            }
            // Min should be 0, max should be 1 (or both 0 if constant)
            prop_assert!(min_val >= -1e-4, "Column {} min should be >= 0, got {}", j, min_val);
            prop_assert!(max_val <= 1.0 + 1e-4, "Column {} max should be <= 1, got {}", j, max_val);
        }
    }

    #[test]
    fn minmax_scaler_inverse_recovers_data(data in matrix_strategy(8, 2)) {
        let mut scaler = MinMaxScaler::new();
        let transformed = scaler.fit_transform(&data).expect("Test data should be valid");
        let recovered = scaler.inverse_transform(&transformed).expect("Test data should be valid");

        let (n_rows, n_cols) = data.shape();
        for i in 0..n_rows {
            for j in 0..n_cols {
                prop_assert!(
                    (data.get(i, j) - recovered.get(i, j)).abs() < 1e-3,
                    "Mismatch at ({}, {}): expected {}, got {}",
                    i, j, data.get(i, j), recovered.get(i, j)
                );
            }
        }
    }

    #[test]
    fn minmax_scaler_custom_range_bounds(data in matrix_strategy(10, 2)) {
        let mut scaler = MinMaxScaler::new().with_range(-1.0, 1.0);
        let transformed = scaler.fit_transform(&data).expect("Test data should be valid");

        let (n_rows, n_cols) = transformed.shape();
        for j in 0..n_cols {
            let mut min_val = f32::INFINITY;
            let mut max_val = f32::NEG_INFINITY;
            for i in 0..n_rows {
                let val = transformed.get(i, j);
                if val < min_val {
                    min_val = val;
                }
                if val > max_val {
                    max_val = val;
                }
            }
            // Min should be -1, max should be 1 (or both -1 if constant)
            prop_assert!(min_val >= -1.0 - 1e-4, "Column {} min should be >= -1, got {}", j, min_val);
            prop_assert!(max_val <= 1.0 + 1e-4, "Column {} max should be <= 1, got {}", j, max_val);
        }
    }

}