Incorporate pr comments

codecov.yml

@@ -22,4 +22,4 @@ ignore:
   - "src/estimagic/optimization/subsolvers/bounded_newton_quadratic_fast.py"
   - "src/estimagic/optimization/subsolvers/_trsbox_quadratic_fast.py"
   - "src/estimagic/optimization/subsolvers/_conjugate_gradient_quadratic_fast.py"
-  - "src/estimagic/optimization/subsolvers/_steinhaug_toint_quadratic_fast.py"
+  - "src/estimagic/optimization/subsolvers/_steihaug_toint_quadratic_fast.py"

src/estimagic/optimization/subsolvers/_conjugate_gradient_quadratic_fast.py

@@ -1,6 +1,4 @@
 """Implementation of the Conjugate Gradient algorithm."""
-import math
-
 import numpy as np
 from numba import njit
 
@@ -80,14 +78,13 @@ def _update_vectors_for_next_iteration(
         direction (np.ndarray): Direction vector of shape (n,).
 
     Returns:
-        (tuple) Tuple containing:
-            - x_candidate (np.ndarray): Updated candidate vector of shape (n,).
-            - residual (np.ndarray): Updated array of residuals of shape (n,).
-            - direction (np.darray): Updated direction vector of shape (n,).
+            x_candidate (np.ndarray): Updated candidate vector of shape (n,).
+            residual_new (np.ndarray): Updated array of residuals of shape (n,).
+            direction (np.darray): Updated direction vector of shape (n,).
     """
     residual_new = np.zeros(len(residual))
-    nom = 0
-    denom = 0
+    nom = 0.0
+    denom = 0.0
     for i in range(len(x_candidate)):
         x_candidate[i] = x_candidate[i] + alpha * direction[i]
         temp = 0
@@ -129,7 +126,7 @@ def _get_distance_to_trustregion_boundary(candidate, direction, radius):
         cc += candidate[i] ** 2
         dd += direction[i] ** 2
         cd += candidate[i] * direction[i]
-    sigma = -cd + math.sqrt(cd * cd + dd * (radius**2 - cc))
+    sigma = -cd + np.sqrt(cd * cd + dd * (radius**2 - cc))
     sigma = sigma / dd
 
     return sigma

src/estimagic/optimization/subsolvers/_trsbox_quadratic_fast.py

@@ -150,14 +150,14 @@ def minimize_trust_trsbox_fast(
                 hess_g,
             )
 
-        if index_bound_active.size > 0:
+        if index_bound_active != -1:
             n_fixed_variables += 1
             if gradient_projected[index_bound_active] >= 0:
                 x_bounded[index_bound_active] = 1
             else:
                 x_bounded[index_bound_active] = -1
 
-            delta_sq = (delta_sq - x_candidate[index_bound_active] ** 2)[0]
+            delta_sq = delta_sq - x_candidate[index_bound_active] ** 2
             if delta_sq <= 0:
                 need_alt_trust_step = True
                 break
@@ -415,7 +415,7 @@ def _update_candidate_vectors_and_reduction(
     """Update candidate vectors and the associated criterion reduction."""
     current_min = g_hess_g / gradient_projected_sumsq
 
-    if index_bound_active.size == 0 and current_min > 0:
+    if index_bound_active == -1 and current_min > 0:
         if curve_min != -1.0:
             curve_min = min(curve_min, current_min)
         else:
@@ -451,6 +451,7 @@ def _take_constrained_step_up_to_boundary(
     x_candidate, gradient_projected, step_len, lower_bounds, upper_bounds
 ):
     """Reduce step length, where boundary is hit, to preserve simple bounds."""
+    index_bound_active = -1
     for i in range(len(x_candidate)):
         if gradient_projected[i] != 0:
             if gradient_projected[i] > 0:
@@ -463,7 +464,7 @@ def _take_constrained_step_up_to_boundary(
                 ) / gradient_projected[i]
             if step_len_constr < step_len:
                 step_len = step_len_constr
-                index_bound_active = np.array([i])
+                index_bound_active = i
 
     return step_len, index_bound_active
 

tests/optimization/tranquilo/test_minimize_trust_region.py

@@ -187,7 +187,7 @@ def test_update_candidate_vector_and_reduction():
     step_len = 0.05
     total_reduction = 0
     curve_min = -0.5
-    index_bound_active = np.array([3, 4])
+    index_bound_active = 3
     gradient_projected_sumsq = 25
     gradient_sumsq = 25
     g_hess_g = 100