Change ns and xs to be a vec instead of array

src/fitter/recursive.rs

@@ -14,8 +14,8 @@ pub struct RecursiveFitter {
     ts: Vec<f64>,
     ms: Vec<f64>,    // TODO Replace with a vec
     vs: Array1<f64>, // TODO Replace with a vec
-    ns: ArrayD<f64>, // TODO Replace with a vec
-    xs: ArrayD<f64>, // TODO Replace with a vec
+    ns: Vec<f64>,    // TODO Replace with a vec
+    xs: Vec<f64>,    // TODO Replace with a vec
     is_fitted: bool,
     h: Array1<f64>,
     i: Array2<f64>,
@@ -40,8 +40,8 @@ impl RecursiveFitter {
             ts: Vec::new(),
             ms: Vec::new(),
             vs: Array1::zeros(0),
-            ns: Array::zeros(0).into_dyn(),
-            xs: Array::zeros(0).into_dyn(),
+            ns: Vec::new(),
+            xs: Vec::new(),
             is_fitted: true,
             h,
             i: Array::eye(m),
@@ -99,13 +99,11 @@ impl Fitter for RecursiveFitter {
         }
 
         // Usual variables.
-        let zeros = Array::zeros(n_new).into_dyn();
-
         self.ts.extend(self.ts_new.iter());
         self.ms.extend(iter::repeat(0.0).take(n_new));
         self.vs = stack![Axis(0), self.vs, self.kernel.k_diag(&self.ts_new)];
-        self.ns = stack![Axis(0), self.ns, zeros];
-        self.xs = stack![Axis(0), self.xs, zeros];
+        self.ns.extend(iter::repeat(0.0).take(n_new));
+        self.xs.extend(iter::repeat(0.0).take(n_new));
 
         // Initialize the predictive, filtering and smoothing distributions.
         for t in &self.ts_new {