Add looped version of gradient, next step would be to jit the function.

src/calviper/math/optimizer.py

@@ -1,4 +1,5 @@
 import itertools
+import numba
 
 import numpy as np
 
@@ -30,6 +31,7 @@ def gradient_(target: np.ndarray, model: np.ndarray, parameter: np.ndarray) -> n
         return gradient_
 
     @staticmethod
+    #@numba.njit
     def gradient(target: np.ndarray, model: np.ndarray, parameter: np.ndarray) -> np.ndarray:
         # cache_ = target, model.conj()
         # numerator_ = np.matmul(cache_, parameter)
@@ -38,8 +40,8 @@ def gradient(target: np.ndarray, model: np.ndarray, parameter: np.ndarray) -> np
         target = target.reshape(n_time, n_channel, 2, 2, n_antennas, n_antennas)
         model = model.reshape(n_time, n_channel, 2, 2, n_antennas, n_antennas)
 
-        numerator_ = np.zeros((n_time, n_channel, 2, n_antennas))
-        denominator_ = np.zeros((n_time, n_channel, 2, n_antennas))
+        numerator_ = np.zeros((n_time, n_channel, 2, n_antennas), dtype=np.complex64)
+        denominator_ = np.zeros((n_time, n_channel, 2, n_antennas), dtype=np.complex64)
 
         # polarizations per baseline are in the order [XX, XY, YX, YY] I think ... so
         for p, q in itertools.product([0, 1], [0, 1]):

src/calviper/math/solver/least_squares.py

@@ -146,8 +146,8 @@ def solve(self, vis, iterations, optimizer=MeanSquaredError(), stopping=1e-3):
 
             self.losses.append(optimizer.loss(y_pred, vis))
 
-            #if n % (iterations // 10) == 0:
-            #    logger.info(f"iteration: {n}\tloss: {np.abs(self.losses[-1])}")
+            if n % (iterations // 10) == 0:
+                logger.info(f"iteration: {n}\tloss: {np.abs(self.losses[-1])}")
 
             if self.losses[-1] < stopping:
                 logger.info(f"Iteration: ({n})\tStopping criterion reached: {self.losses[-1]}")