Unify optimizer parameters (qiskit-community#1084)

qiskit/aqua/components/optimizers/gsls.py

@@ -14,6 +14,7 @@
 
 """Line search with Gaussian-smoothed samples on a sphere."""
 
+import warnings
 from typing import Dict, Optional, Tuple, List, Callable
 import logging
 import numpy as np
@@ -32,14 +33,14 @@ class GSLS(Optimizer):
     based on Gaussian-smoothed samples on a sphere.
     """
 
-    _OPTIONS = ['max_iter', 'max_eval', 'disp', 'sampling_radius',
+    _OPTIONS = ['maxiter', 'max_eval', 'disp', 'sampling_radius',
                 'sample_size_factor', 'initial_step_size', 'min_step_size',
                 'step_size_multiplier', 'armijo_parameter',
                 'min_gradient_norm', 'max_failed_rejection_sampling']
 
     # pylint:disable=unused-argument
     def __init__(self,
-                 max_iter: int = 10000,
+                 maxiter: int = 10000,
                  max_eval: int = 10000,
                  disp: bool = False,
                  sampling_radius: float = 1.0e-6,
@@ -49,10 +50,11 @@ def __init__(self,
                  step_size_multiplier: float = 0.4,
                  armijo_parameter: float = 1.0e-1,
                  min_gradient_norm: float = 1e-8,
-                 max_failed_rejection_sampling: int = 50) -> None:
+                 max_failed_rejection_sampling: int = 50,
+                 max_iter: Optional[int] = None) -> None:
         """
         Args:
-            max_iter: Maximum number of iterations.
+            maxiter: Maximum number of iterations.
             max_eval: Maximum number of evaluations.
             disp: Set to True to display convergence messages.
             sampling_radius: Sampling radius to determine gradient estimate.
@@ -67,8 +69,15 @@ def __init__(self,
             min_gradient_norm: If the gradient norm is below this threshold, the algorithm stops.
             max_failed_rejection_sampling: Maximum number of attempts to sample points within
                 bounds.
+            max_iter: Deprecated, use maxiter.
         """
         super().__init__()
+        if max_iter is not None:
+            warnings.warn('The max_iter parameter is deprecated as of '
+                          '0.8.0 and will be removed no sooner than 3 months after the release. '
+                          'You should use maxiter instead.',
+                          DeprecationWarning)
+            maxiter = max_iter
         for k, v in locals().items():
             if k in self._OPTIONS:
                 self._options[k] = v
@@ -153,7 +162,7 @@ def ls_optimize(self, n: int, obj_fun: Callable, initial_point: np.ndarray, var_
         x = initial_point
         x_value = obj_fun(x)
         n_evals += 1
-        while iter_count < self._options['max_iter'] \
+        while iter_count < self._options['maxiter'] \
                 and n_evals < self._options['max_eval']:
 
             # Determine set of sample points

qiskit/aqua/components/optimizers/spsa.py

@@ -14,6 +14,8 @@
 
 """Simultaneous Perturbation Stochastic Approximation optimizer."""
 
+import warnings
+from typing import Optional, List, Callable
 import logging
 
 import numpy as np
@@ -62,18 +64,19 @@ class SPSA(Optimizer):
 
     # pylint: disable=unused-argument
     def __init__(self,
-                 max_trials: int = 1000,
+                 maxiter: int = 1000,
                  save_steps: int = 1,
                  last_avg: int = 1,
                  c0: float = _C0,
                  c1: float = 0.1,
                  c2: float = 0.602,
                  c3: float = 0.101,
                  c4: float = 0,
-                 skip_calibration: float = False) -> None:
+                 skip_calibration: float = False,
+                 max_trials: Optional[int] = None) -> None:
         """
         Args:
-            max_trials: Maximum number of iterations to perform.
+            maxiter: Maximum number of iterations to perform.
             save_steps: Save intermediate info every save_steps step. It has a min. value of 1.
             last_avg: Averaged parameters over the last_avg iterations.
                 If last_avg = 1, only the last iteration is considered. It has a min. value of 1.
@@ -83,14 +86,21 @@ def __init__(self,
             c3: The gamma in the paper, and it is used to adjust c (c1) at each iteration.
             c4: The parameter used to control a as well.
             skip_calibration: Skip calibration and use provided c(s) as is.
+            max_trials: Deprecated, use maxiter.
         """
         validate_min('save_steps', save_steps, 1)
         validate_min('last_avg', last_avg, 1)
         super().__init__()
+        if max_trials is not None:
+            warnings.warn('The max_trials parameter is deprecated as of '
+                          '0.8.0 and will be removed no sooner than 3 months after the release. '
+                          'You should use maxiter instead.',
+                          DeprecationWarning)
+            maxiter = max_trials
         for k, v in locals().items():
             if k in self._OPTIONS:
                 self._options[k] = v
-        self._max_trials = max_trials
+        self._maxiter = maxiter
         self._parameters = np.array([c0, c1, c2, c3, c4])
         self._skip_calibration = skip_calibration
 
@@ -113,33 +123,37 @@ def optimize(self, num_vars, objective_function, gradient_function=None,
         logger.debug('Parameters: %s', self._parameters)
         if not self._skip_calibration:
             # at least one calibration, at most 25 calibrations
-            num_steps_calibration = min(25, max(1, self._max_trials // 5))
+            num_steps_calibration = min(25, max(1, self._maxiter // 5))
             self._calibration(objective_function, initial_point, num_steps_calibration)
         else:
             logger.debug('Skipping calibration, parameters used as provided.')
 
         opt, sol, _, _, _, _ = self._optimization(objective_function,
                                                   initial_point,
-                                                  max_trials=self._max_trials,
+                                                  maxiter=self._maxiter,
                                                   **self._options)
         return sol, opt, None
 
-    def _optimization(self, obj_fun, initial_theta, max_trials, save_steps=1, last_avg=1):
+    def _optimization(self,
+                      obj_fun: Callable,
+                      initial_theta: np.ndarray,
+                      maxiter: int,
+                      save_steps: int = 1,
+                      last_avg: int = 1) -> List:
         """Minimizes obj_fun(theta) with a simultaneous perturbation stochastic
         approximation algorithm.
 
         Args:
-            obj_fun (callable): the function to minimize
-            initial_theta (numpy.array): initial value for the variables of
-                obj_fun
-            max_trials (int) : the maximum number of trial steps ( = function
+            obj_fun: the function to minimize
+            initial_theta: initial value for the variables of obj_fun
+            maxiter: the maximum number of trial steps ( = function
                 calls/2) in the optimization
-            save_steps (int) : stores optimization outcomes each 'save_steps'
+            save_steps: stores optimization outcomes each 'save_steps'
                 trial steps
-            last_avg (int) : number of last updates of the variables to average
+            last_avg: number of last updates of the variables to average
                 on for the final obj_fun
         Returns:
-            list: a list with the following elements:
+            a list with the following elements:
                 cost_final : final optimized value for obj_fun
                 theta_best : final values of the variables corresponding to
                     cost_final
@@ -159,7 +173,7 @@ def _optimization(self, obj_fun, initial_theta, max_trials, save_steps=1, last_a
         cost_minus_save = []
         theta = initial_theta
         theta_best = np.zeros(initial_theta.shape)
-        for k in range(max_trials):
+        for k in range(maxiter):
             # SPSA Parameters
             a_spsa = float(self._parameters[0]) / np.power(k + 1 + self._parameters[4],
                                                            self._parameters[2])
@@ -187,7 +201,7 @@ def _optimization(self, obj_fun, initial_theta, max_trials, save_steps=1, last_a
                 cost_plus_save.append(cost_plus)
                 cost_minus_save.append(cost_minus)
 
-            if k >= max_trials - last_avg:
+            if k >= maxiter - last_avg:
                 theta_best += theta / last_avg
         # final cost update
         cost_final = obj_fun(theta_best)
@@ -196,7 +210,10 @@ def _optimization(self, obj_fun, initial_theta, max_trials, save_steps=1, last_a
         return [cost_final, theta_best, cost_plus_save, cost_minus_save,
                 theta_plus_save, theta_minus_save]
 
-    def _calibration(self, obj_fun, initial_theta, stat):
+    def _calibration(self,
+                     obj_fun: Callable,
+                     initial_theta: np.ndarray,
+                     stat: int):
         """Calibrates and stores the SPSA parameters back.
 
         SPSA parameters are c0 through c5 stored in parameters array
@@ -207,11 +224,9 @@ def _calibration(self, obj_fun, initial_theta, stat):
         c1 is initial_c and is first perturbation of initial_theta.
 
         Args:
-            obj_fun (callable): the function to minimize.
-            initial_theta (numpy.array): initial value for the variables of
-                obj_fun.
-            stat (int) : number of random gradient directions to average on in
-                the calibration.
+            obj_fun: the function to minimize.
+            initial_theta: initial value for the variables of obj_fun.
+            stat: number of random gradient directions to average on in the calibration.
         """
 
         target_update = self._parameters[0]

qiskit/aqua/utils/qp_solver.py

@@ -14,6 +14,7 @@
 
 """ qp solver """
 
+import warnings
 from typing import Optional, Tuple
 import logging
 
@@ -30,8 +31,9 @@
 def optimize_svm(kernel_matrix: np.ndarray,
                  y: np.ndarray,
                  scaling: Optional[float] = None,
-                 max_iters: int = 500,
-                 show_progress: bool = False) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
+                 maxiter: int = 500,
+                 show_progress: bool = False,
+                 max_iters: Optional[int] = None) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
     """
     Solving quadratic programming problem for SVM; thus, some constraints are fixed.
 
@@ -40,8 +42,9 @@ def optimize_svm(kernel_matrix: np.ndarray,
         y: Nx1 array
         scaling: the scaling factor to renormalize the `y`, if it is None,
                  use L2-norm of `y` for normalization
-        max_iters: number of iterations for QP solver
+        maxiter: number of iterations for QP solver
         show_progress: showing the progress of QP solver
+        max_iters: Deprecated, use maxiter.
 
     Returns:
         np.ndarray: Sx1 array, where S is the number of supports
@@ -56,6 +59,12 @@ def optimize_svm(kernel_matrix: np.ndarray,
         raise NameError("The CVXPY package is required to use the "
                         "optimize_svm() function. You can install it with "
                         "'pip install qiskit-aqua[cvx]'.")
+    if max_iters is not None:
+        warnings.warn('The max_iters parameter is deprecated as of '
+                      '0.8.0 and will be removed no sooner than 3 months after the release. '
+                      'You should use maxiter instead.',
+                      DeprecationWarning)
+        maxiter = max_iters
     if y.ndim == 1:
         y = y[:, np.newaxis]
     H = np.outer(y, y) * kernel_matrix

qiskit/optimization/algorithms/admm_optimizer.py

@@ -13,6 +13,7 @@
 # that they have been altered from the originals.
 
 """An implementation of the ADMM algorithm."""
+import warnings
 import copy
 import logging
 import time
@@ -37,18 +38,27 @@
 class ADMMParameters:
     """Defines a set of parameters for ADMM optimizer."""
 
-    def __init__(self, rho_initial: float = 10000, factor_c: float = 100000, beta: float = 1000,
-                 max_iter: int = 10, tol: float = 1.e-4, max_time: float = np.inf,
-                 three_block: bool = True, vary_rho: int = UPDATE_RHO_BY_TEN_PERCENT,
-                 tau_incr: float = 2, tau_decr: float = 2, mu_res: float = 10,
-                 mu_merit: float = 1000) -> None:
+    def __init__(self,
+                 rho_initial: float = 10000,
+                 factor_c: float = 100000,
+                 beta: float = 1000,
+                 maxiter: int = 10,
+                 tol: float = 1.e-4,
+                 max_time: float = np.inf,
+                 three_block: bool = True,
+                 vary_rho: int = UPDATE_RHO_BY_TEN_PERCENT,
+                 tau_incr: float = 2,
+                 tau_decr: float = 2,
+                 mu_res: float = 10,
+                 mu_merit: float = 1000,
+                 max_iter: Optional[int] = None) -> None:
         """Defines parameters for ADMM optimizer and their default values.
 
         Args:
             rho_initial: Initial value of rho parameter of ADMM.
             factor_c: Penalizing factor for equality constraints, when mapping to QUBO.
             beta: Penalization for y decision variables.
-            max_iter: Maximum number of iterations for ADMM.
+            maxiter: Maximum number of iterations for ADMM.
             tol: Tolerance for the residual convergence.
             max_time: Maximum running time (in seconds) for ADMM.
             three_block: Boolean flag to select the 3-block ADMM implementation.
@@ -64,8 +74,15 @@ def __init__(self, rho_initial: float = 10000, factor_c: float = 100000, beta: f
             tau_decr: Parameter used in the rho update (UPDATE_RHO_BY_RESIDUALS).
             mu_res: Parameter used in the rho update (UPDATE_RHO_BY_RESIDUALS).
             mu_merit: Penalization for constraint residual. Used to compute the merit values.
+            max_iter: Deprecated, use maxiter.
         """
         super().__init__()
+        if max_iter is not None:
+            warnings.warn('The max_iter parameter is deprecated as of '
+                          '0.8.0 and will be removed no sooner than 3 months after the release. '
+                          'You should use maxiter instead.',
+                          DeprecationWarning)
+            maxiter = max_iter
         self.mu_merit = mu_merit
         self.mu_res = mu_res
         self.tau_decr = tau_decr
@@ -74,7 +91,7 @@ def __init__(self, rho_initial: float = 10000, factor_c: float = 100000, beta: f
         self.three_block = three_block
         self.max_time = max_time
         self.tol = tol
-        self.max_iter = max_iter
+        self.maxiter = maxiter
         self.factor_c = factor_c
         self.beta = beta
         self.rho_initial = rho_initial
@@ -278,7 +295,7 @@ def solve(self, problem: QuadraticProgram) -> ADMMOptimizationResult:
         iteration = 0
         residual = 1.e+2
 
-        while (iteration < self._params.max_iter and residual > self._params.tol) \
+        while (iteration < self._params.maxiter and residual > self._params.tol) \
                 and (elapsed_time < self._params.max_time):
             if self._state.step1_absolute_indices:
                 op1 = self._create_step1_problem()

releasenotes/notes/unify-optimizer-params-a73247cd97065bc6.yaml

@@ -0,0 +1,7 @@
+---
+deprecations:
+  - |
+    GSLS optimizer class deprecated ``__init__`` parameter ``max_iter`` in favor of ``maxiter``.
+    SPSA optimizer class deprecated ``__init__`` parameter ``max_trials`` in favor of ``maxiter``.
+    optimize_svm function deprecated ``max_iters`` parameter in favor of ``maxiter``.
+    ADMMParameters class deprecated ``__init__`` parameter ``max_iter`` in favor of ``maxiter``.

test/aqua/test_optimizers.py

@@ -89,7 +89,7 @@ def test_slsqp(self):
     @unittest.skip("Skipping SPSA as it does not do well on non-convex rozen")
     def test_spsa(self):
         """ spsa test """
-        optimizer = SPSA(max_trials=10000)
+        optimizer = SPSA(maxiter=10000)
         res = self._optimize(optimizer)
         self.assertLessEqual(res[2], 100000)
 
@@ -101,7 +101,7 @@ def test_tnc(self):
 
     def test_gsls(self):
         """ gsls test """
-        optimizer = GSLS(sample_size_factor=40, sampling_radius=1.0e-12, max_iter=10000,
+        optimizer = GSLS(sample_size_factor=40, sampling_radius=1.0e-12, maxiter=10000,
                          max_eval=10000, min_step_size=1.0e-12)
         x_0 = [1.3, 0.7, 0.8, 1.9, 1.2]
         _, x_value, n_evals = optimizer.optimize(len(x_0), rosen, initial_point=x_0)

test/aqua/test_vqc.py

@@ -69,7 +69,7 @@ def setUp(self):
                                               seed_simulator=self.seed,
                                               seed_transpiler=self.seed)
 
-        self.spsa = SPSA(max_trials=10, save_steps=1, c0=4.0, c1=0.1, c2=0.602, c3=0.101,
+        self.spsa = SPSA(maxiter=10, save_steps=1, c0=4.0, c1=0.1, c2=0.602, c3=0.101,
                          c4=0.0, skip_calibration=True)
 
     def assertSimpleClassificationIsCorrect(self, vqc, backend=None, ref_opt_params=None,

test/aqua/test_vqe.py

@@ -120,7 +120,7 @@ def test_missing_varform_params(self):
 
     @data(
         (SLSQP(maxiter=50), 5, 4),
-        (SPSA(max_trials=150), 3, 2),  # max_evals_grouped=n or =2 if n>2
+        (SPSA(maxiter=150), 3, 2),  # max_evals_grouped=n or =2 if n>2
     )
     @unpack
     def test_max_evals_grouped(self, optimizer, places, max_evals_grouped):
@@ -133,7 +133,7 @@ def test_max_evals_grouped(self, optimizer, places, max_evals_grouped):
 
     def test_basic_aer_qasm(self):
         """Test the VQE on BasicAer's QASM simulator."""
-        optimizer = SPSA(max_trials=300, last_avg=5)
+        optimizer = SPSA(maxiter=300, last_avg=5)
         wavefunction = self.ry_wavefunction
 
         vqe = VQE(self.h2_op, wavefunction, optimizer, max_evals_grouped=1)

test/aqua/test_vqe2iqpe.py

@@ -48,7 +48,7 @@ def test_vqe_2_iqpe(self):
         num_qbits = self.qubit_op.num_qubits
         wavefunction = TwoLocal(num_qbits, ['ry', 'rz'], 'cz', reps=3, insert_barriers=True)
 
-        optimizer = SPSA(max_trials=10)
+        optimizer = SPSA(maxiter=10)
         algo = VQE(self.qubit_op, wavefunction, optimizer)
 
         quantum_instance = QuantumInstance(backend, seed_simulator=self.seed,