ADAM bug when calculating the gradient in batches #178 (#183)

Co-authored-by: Peter Röseler <peter.roeseler@gmail.com>
Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com>
Co-authored-by: Elena Peña Tapia <57907331+ElePT@users.noreply.github.com>
(cherry picked from commit 6724b47)

qiskit_algorithms/optimizers/adam_amsgrad.py

@@ -1,6 +1,6 @@
 # This code is part of a Qiskit project.
 #
-# (C) Copyright IBM 2019, 2023.
+# (C) Copyright IBM 2019, 2024.
 #
 # This code is licensed under the Apache License, Version 2.0. You may
 # obtain a copy of this license in the LICENSE.txt file in the root directory
@@ -209,7 +209,9 @@ def minimize(
  The result of the optimization, containing e.g. the result as attribute ``x``.
  """
  if jac is None:
- jac = Optimizer.wrap_function(Optimizer.gradient_num_diff, (fun, self._eps))
+ jac = Optimizer.wrap_function(
+ Optimizer.gradient_num_diff, (fun, self._eps, self._max_evals_grouped)
+ )
 
  derivative = jac(x0)
  self._t = 0

releasenotes/notes/fix_aqgd_max_grouped_evals-fbe108c005a9b7ac.yaml

@@ -0,0 +1,9 @@
+---
+fixes:
+ - |
+ Fixed the AQGD optimizer grouping objective function calls by default so that a single point is now passed to the
+ objective function. For algorithms that can handle more than one gradient evaluations in their objective function,
+ such as a VQE in the algorithms here, the number of grouped evaluations can be controlled via the max_grouped_evals
+ parameter. Grouped evaluations allows a list of points to be handed over so that they can potentially be assessed
+ more efficiently in a single job.
+

test/optimizers/test_adam.py

@@ -0,0 +1,86 @@
+# This code is part of a Qiskit project.
+#
+# (C) Copyright IBM 2024.
+#
+# This code is licensed under the Apache License, Version 2.0. You may
+# obtain a copy of this license in the LICENSE.txt file in the root directory
+# of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.
+#
+# Any modifications or derivative works of this code must retain this
+# copyright notice, and modified files need to carry a notice indicating
+# that they have been altered from the originals.
+
+"""Tests for the ADAM optimizer."""
+
+from test import QiskitAlgorithmsTestCase
+
+from ddt import ddt, data
+import numpy as np
+
+from qiskit_algorithms.optimizers import ADAM, Optimizer
+from qiskit_algorithms.utils import algorithm_globals
+
+
+@ddt
+class TestADAM(QiskitAlgorithmsTestCase):
+ """Tests for the ADAM optimizer."""
+
+ def setUp(self):
+ super().setUp()
+ algorithm_globals.random_seed = 52
+ # Feature vector
+ self.x = np.array([1, 2, 3, 4])
+ # Target value
+ self.y = 5
+
+ def objective(self, w):
+ """
+ Objective function to minimize mean squared error.
+
+ Parameters:
+ w : numpy array
+ The weights (including bias) for the linear model.
+
+ Returns:
+ float
+ The mean squared error.
+ """
+ # Extract weights and bias from the parameter vector
+ new_shape = (5, int(len(w) / 5))
+ w = np.reshape(w, new_shape)
+
+ weights = w[:-1, :]
+ bias = w[-1, :]
+ # Calculate the predicted values
+ y_pred = np.dot(self.x, weights) + bias
+ # Calculate the mean squared error
+ mse = np.mean((self.y - y_pred) ** 2)
+ return mse
+
+ def run_optimizer(self, optimizer: Optimizer, weights: np.ndarray, max_nfev: int):
+ """Test the optimizer.
+
+ Args:
+ optimizer: The optimizer instance to test.
+ weights: The weights to optimize.
+ max_nfev: The maximal allowed number of function evaluations.
+ """
+
+ # Minimize
+ res = optimizer.minimize(self.objective, np.array(weights), None)
+ error = res.fun
+ nfev = res.nfev
+
+ self.assertAlmostEqual(error, 0, places=3)
+ self.assertLessEqual(nfev, max_nfev)
+
+ @data(1, 5)
+ def test_adam_max_evals(self, max_evals_grouped):
+ """adam test"""
+ # Initialize weights (including bias)
+ w = np.zeros(len(self.x) + 1)
+ # Initialize optimizer
+ optimizer = ADAM(maxiter=10000, tol=1e-06)
+ # Test one evaluation at a time
+ optimizer.set_max_evals_grouped(max_evals_grouped)
+ self.run_optimizer(optimizer, w, max_nfev=10000)