Add MP test

Author: bytesnake

ndarray-linalg/Cargo.toml

@@ -50,7 +50,8 @@ paste = "1.0"
 criterion = "0.3"
 # Keep the same version as ndarray's dependency!
 approx = { version = "0.4", features = ["num-complex"] }
-rand_xoshiro = "0.6"
+rand_xoshiro = "0.4"
+ndarray-rand = "0.12"
 
 [[bench]]
 name = "truncated_eig"

ndarray-linalg/src/lobpcg/eig.rs

@@ -20,14 +20,17 @@ use num_traits::{Float, NumCast};
 /// # Example
 ///
 /// ```rust
+/// use ndarray::{arr1, Array2};
+/// use ndarray_linalg::{TruncatedEig, TruncatedOrder};
+///
 /// let diag = arr1(&[1., 2., 3., 4., 5.]);
 /// let a = Array2::from_diag(&diag);
 ///
-/// let eig = TruncatedEig::new(a, Order::Largest)
+/// let eig = TruncatedEig::new(a, TruncatedOrder::Largest)
 ///    .precision(1e-5)
 ///    .maxiter(500);
 ///
-/// let res = eig.decompose();
+/// let res = eig.decompose(3);
 /// ```
 
 pub struct TruncatedEig<A: Scalar> {
@@ -109,14 +112,17 @@ impl<A: Float + Scalar + ScalarOperand + Lapack + PartialOrd + Default> Truncate
     /// # Example
     ///
     /// ```rust
+    /// use ndarray::{arr1, Array2};
+    /// use ndarray_linalg::{TruncatedEig, TruncatedOrder};
+    ///
     /// let diag = arr1(&[1., 2., 3., 4., 5.]);
     /// let a = Array2::from_diag(&diag);
     ///
-    /// let eig = TruncatedEig::new(a, Order::Largest)
+    /// let eig = TruncatedEig::new(a, TruncatedOrder::Largest)
     ///    .precision(1e-5)
     ///    .maxiter(500);
     ///
-    /// let res = eig.decompose();
+    /// let res = eig.decompose(3);
     /// ```
     pub fn decompose(&self, num: usize) -> LobpcgResult<A> {
         let x: Array2<f64> = generate::random((self.problem.len_of(Axis(0)), num));
@@ -169,15 +175,21 @@ impl<A: Float + Scalar + ScalarOperand + Lapack + PartialOrd + Default> IntoIter
 /// # Example
 ///
 /// ```rust
-/// let teig = TruncatedEig::new(a, Order::Largest)
+/// use ndarray::{arr1, Array2};
+/// use ndarray_linalg::{TruncatedEig, TruncatedOrder};
+///
+/// let diag = arr1(&[1., 2., 3., 4., 5.]);
+/// let a = Array2::from_diag(&diag);
+///
+/// let teig = TruncatedEig::new(a, TruncatedOrder::Largest)
 ///     .precision(1e-5)
 ///     .maxiter(500);
 /// 
 /// // solve eigenproblem until eigenvalues get smaller than 0.5
 /// let res = teig.into_iter()
 ///     .take_while(|x| x.0[0] > 0.5)
-///     .flat_map(|x| x.0)
-///     .collect();
+///     .flat_map(|x| x.0.to_vec())
+///     .collect::<Vec<_>>();
 /// ```
 pub struct TruncatedEigIterator<A: Scalar> {
     step_size: usize,

ndarray-linalg/src/lobpcg/svd.rs

@@ -200,8 +200,11 @@ mod tests {
     use super::TruncatedSvd;
     use crate::{close_l2, generate};
 
-    use ndarray::{arr1, arr2, Array2};
+    use rand::SeedableRng;
     use rand_xoshiro::Xoshiro256Plus;
+    use ndarray::{arr1, arr2, Array1, Array2};
+    use ndarray_rand::{rand_distr::StandardNormal, RandomExt};
+
     use approx::assert_abs_diff_eq;
 
     #[test]
@@ -246,14 +249,16 @@ mod tests {
     #[test]
     fn test_marchenko_pastur() {
         // create random number generator
-        let mut rng = SmallRng::seed_from_u64(3);
+        let mut rng = Xoshiro256Plus::seed_from_u64(3);
 
         // generate normal distribution random data with N >> p
-        let data = Array2::random_using((1000, 500), StandardNormal, &mut rng);
-        let dataset = Dataset::from(data / 1000f64.sqrt());
+        let data = Array2::random_using((1000, 500), StandardNormal, &mut rng) / 1000f64.sqrt();
+
+        let res = TruncatedSvd::new(data, Order::Largest)
+            .decompose(500)
+            .unwrap();
 
-        let model = Pca::params(500).fit(&dataset);
-        let sv = model.singular_values().mapv(|x| x * x); 
+        let sv = res.values().mapv(|x: f64| x*x);
 
         // we have created a random spectrum and can apply the Marchenko-Pastur law
         // with variance 1 and p/n = 0.5