Fix numpy scalar extraction (#372)

* extract numpy scalars properly

* change to one-liner

* Update src/squlearn/optimizers/approximated_gradients.py

Co-authored-by: Florian Wieland <114916947+ProfessorNova@users.noreply.github.com>

* Update src/squlearn/optimizers/approximated_gradients.py

Co-authored-by: Florian Wieland <114916947+ProfessorNova@users.noreply.github.com>

* Update src/squlearn/optimizers/approximated_gradients.py

Co-authored-by: Florian Wieland <114916947+ProfessorNova@users.noreply.github.com>

---------

Co-authored-by: Florian Wieland <114916947+ProfessorNova@users.noreply.github.com>

Author: MoritzWillmann

src/squlearn/kernel/lowlevel_kernel/projected_quantum_kernel.py

@@ -961,10 +961,8 @@ def __call__(
         param = parameters[: qnn.num_parameters]
         param_op = parameters[qnn.num_parameters :]
 
-        if len(param.shape) == 1 and len(param) == 1:
-            param = float(param)
-        if len(param_op.shape) == 1 and len(param_op) == 1:
-            param_op = float(param_op)
+        param = param.item() if param.size == 1 else param
+        param_op = param_op.item() if param_op.size == 1 else param_op
 
         x_result = qnn.evaluate(x, param, param_op, "f")["f"]
         if y is not None:

src/squlearn/optimizers/approximated_gradients.py

@@ -66,33 +66,34 @@ def gradient(self, x: np.ndarray) -> np.ndarray:
             g = np.zeros(len(x))
             for i in range(len(x)):
                 dx = np.eye(1, len(x), k=i)[0] * self.eps
-                g[i] = ((self.fun(x + dx) - f0)) / self.eps  # Extract scalar
+                g[i] = ((self.fun(x + dx) - f0) / self.eps).item()
 
         elif self.formula == "backwards":
             f0 = self.fun(x)
             g = np.zeros(len(x))
             for i in range(len(x)):
                 dx = np.eye(1, len(x), k=i)[0] * self.eps
-                g[i] = ((f0 - self.fun(x - dx))) / self.eps  # Extract scalar
+                g[i] = ((f0 - self.fun(x - dx)) / self.eps).item()
 
         elif self.formula == "central":
             g = np.zeros(len(x))
             for i in range(len(x)):
                 dx = np.eye(1, len(x), k=i)[0] * self.eps
-                g[i] = ((self.fun(x + dx) - self.fun(x - dx))) / (2.0 * self.eps)  # Extract scalar
+                g[i] = ((self.fun(x + dx) - self.fun(x - dx)) / (2.0 * self.eps)).item()
 
         elif self.formula == "five-point":
             g = np.zeros(len(x))
             for i in range(len(x)):
                 dx = np.eye(1, len(x), k=i)[0] * self.eps
                 g[i] = (
-                    -1.0 * self.fun(x + 2.0 * dx)
-                    + 8.0 * self.fun(x + 1.0 * dx)
-                    - 8.0 * self.fun(x - 1.0 * dx)
-                    + 1.0 * self.fun(x - 2.0 * dx)
-                ) / (
-                    12.0 * self.eps
-                )  # Extract scalar
+                    (
+                        -1.0 * self.fun(x + 2.0 * dx)
+                        + 8.0 * self.fun(x + 1.0 * dx)
+                        - 8.0 * self.fun(x - 1.0 * dx)
+                        + 1.0 * self.fun(x - 2.0 * dx)
+                    )
+                    / (12.0 * self.eps)
+                ).item()
         else:
             raise ValueError("Wrong value of type: " + self.formula)