Log calibrated learning rate and perturbation in (QN-)SPSA (Qiskit#7109)

* log calibrated learning rate and perturbation * Fix formatting Co-authored-by: Jake Lishman <[email protected]>
ElePT · Oct 8, 2021 · dce862f · dce862f
1 parent 9a376a5
commit dce862f
Showing 1 changed file with 12 additions and 4 deletions.
diff --git a/qiskit/algorithms/optimizers/spsa.py b/qiskit/algorithms/optimizers/spsa.py
@@ -315,6 +315,7 @@ def calibrate(
             tuple(generator, generator): A tuple of powerseries generators, the first one for the
                 learning rate and the second one for the perturbation.
         """
+        logger.info("SPSA: Starting calibration of learning rate and perturbation.")
         if target_magnitude is None:
             target_magnitude = 2 * np.pi / 10
 
@@ -347,6 +348,15 @@ def calibrate(
             warnings.warn(f"Calibration failed, using {target_magnitude} for `a`")
             a = target_magnitude
 
+        logger.info("Finished calibration:")
+        logger.info(
+            " -- Learning rate: a / ((A + n) ^ alpha) with a = %s, A = %s, alpha = %s",
+            a,
+            stability_constant,
+            alpha,
+        )
+        logger.info(" -- Perturbation: c / (n ^ gamma) with c = %s, gamma = %s", c, gamma)
+
         # set up the powerseries
         def learning_rate():
             return powerseries(a, alpha, stability_constant)
@@ -529,8 +539,7 @@ def minimize(
                     fun, x, max_evals_grouped=self._max_evals_grouped
                 )
 
-        logger.info("=" * 30)
-        logger.info("Starting SPSA optimization")
+        logger.info("SPSA: Starting optimization.")
         start = time()
 
         # keep track of the last few steps to return their average
@@ -611,8 +620,7 @@ def minimize(
                     logger.info("terminated optimization at {k}/{self.maxiter} iterations")
                     break
 
-        logger.info("SPSA finished in %s", time() - start)
-        logger.info("=" * 30)
+        logger.info("SPSA: Finished in %s", time() - start)
 
         if self.last_avg > 1:
             x = np.mean(last_steps, axis=0)