Revert "Allow SNR optimizer to use candidate point in initial array (g…

…wastro#4393)"

This reverts commit 460d2a40c4bdaf1446fd541f9e7666b24c625d35.

The revert is to avoid a bug in the SNR optimizer introduced by this commit
and later fixed properly in gwastro#4561.

bin/pycbc_optimize_snr

@@ -130,28 +130,13 @@ def compute_minus_network_snr_pso(v, *argv, **kwargs):
     return -nsnr_array
 
 
-def normalize_initial_point(initial_point, bounds):
-    return (initial_point - bounds[:,0]) / (bounds[:,1] - bounds[:,0])
-
-def optimize_di(bounds, cli_args, extra_args, initial_point):
-    bounds = numpy.array([
+def optimize_di(bounds, cli_args, extra_args):
+    bounds = [
         bounds['mchirp'],
         bounds['eta'],
         bounds['spin1z'],
         bounds['spin2z']
-    ])
-
-
-    # Currently only implemented for random seed initial array
-    rng = numpy.random.mtrand._rand
-    population_shape=(cli_args.di_popsize, 4)
-    population = rng.uniform(size=population_shape)
-    if cli_args.include_candidate_in_optimizer:
-        # Re-normalize the initial point into the correct range
-        point_init = normalize_initial_point(initial_point, bounds)
-        # add the initial point to the population
-        population = numpy.concatenate((population[:-1], point_init))
-
+    ]
     results = differential_evolution(
         compute_minus_network_snr,
         bounds,
@@ -161,13 +146,12 @@ def optimize_di(bounds, cli_args, extra_args, initial_point):
         mutation=(0.5, 1),
         recombination=0.7,
         callback=callback_func,
-        args=extra_args,
-        init=population
+        args=extra_args
     )
     return results.x
 
 
-def optimize_shgo(bounds, cli_args, extra_args, initial_point): # pylint: disable=unused-argument
+def optimize_shgo(bounds, cli_args, extra_args):
     bounds = [
         bounds['mchirp'],
         bounds['eta'],
@@ -184,48 +168,30 @@ def optimize_shgo(bounds, cli_args, extra_args, initial_point): # pylint: disabl
     )
     return results.x
 
-def normalize_population(population, min_bounds, max_bounds):
-    norm_pop = min_bounds + population * (max_bounds - min_bounds)
-
-    return norm_pop
 
-def optimize_pso(bounds, cli_args, extra_args, initial_point):
+def optimize_pso(bounds, cli_args, extra_args):
     options = {
         'c1': cli_args.pso_c1,
         'c2': cli_args.pso_c2,
         'w': cli_args.pso_w
     }
-    min_bounds = numpy.array([
+    min_bounds = [
         bounds['mchirp'][0],
         bounds['eta'][0],
         bounds['spin1z'][0],
         bounds['spin2z'][0]
-    ])
-    max_bounds = numpy.array([
+    ]
+    max_bounds = [
         bounds['mchirp'][1],
         bounds['eta'][1],
         bounds['spin1z'][1],
         bounds['spin2z'][1]
-    ])
-
-    # Manually generate the initial points, this is the same as the default
-    # method, but allows us to make some modifications
-    population = numpy.random.uniform(
-        low=0.0, high=1.0, size=(cli_args.pso_particles, 4)
-    )
-    population = normalize_population(population, min_bounds, max_bounds)
-
-    if cli_args.include_candidate_in_optimizer:
-        # add the initial point to the population
-        population = numpy.concatenate((population[:-1],
-                                        initial_point))
-
+    ]
     optimizer = ps.single.GlobalBestPSO(
         n_particles=cli_args.pso_particles,
         dimensions=4,
         options=options,
-        bounds=(min_bounds, max_bounds),
-        init_pos=population
+        bounds=(min_bounds, max_bounds)
     )
     _, results = optimizer.optimize(
         compute_minus_network_snr_pso,
@@ -318,14 +284,6 @@ parser.add_argument('--pso-c2', type=float, default=2.0,
 parser.add_argument('--pso-w', type=float, default=0.01,
                     help='Only relevant for --optimizer pso: '
                     'The hyperparameter w: the inertia parameter.')
-parser.add_argument('--include-candidate-in-optimizer', action='store_true',
-                    help='Include parameters of the candidate event in the '
-                    'initialised array for the optimizer. Only relevant for '
-                    '--optimizer pso or differential_evolution')
-parser.add_argument('--seed', type=int, default=42,
-                    help='Seed to supply to the random generation of initial '
-                    'array to pass to the optimizer. Only relevant for '
-                    '--optimizer pso or differential_evolution')
 
 scheme.insert_processing_option_group(parser)
 fft.insert_fft_option_group(parser)
@@ -338,9 +296,6 @@ if args.optimizer == 'pso' and ps == None:
     parser.error('You need to install pyswarms to use the pso optimizer.')
 pycbc.init_logging(args.verbose)
 
-if args.seed:
-    numpy.random.seed(args.seed)
-
 scheme.verify_processing_options(args, parser)
 fft.verify_fft_options(args, parser)
 
@@ -420,27 +375,11 @@ bounds = {
     'spin2z': (minspin2z, maxspin2z)
 }
 
-if args.include_candidate_in_optimizer:
-    # Initial point from found candidate
-    mchirp_init = cv.mchirp_from_mass1_mass2(fp['mass1'][()], fp['mass2'][()])
-    eta_init = cv.eta_from_mass1_mass2(fp['mass1'][()], fp['mass2'][()])
-    spin1z_init = fp['spin1z'][()]
-    spin2z_init = fp['spin2z'][()]
-
-    initial_point = numpy.array([
-        mchirp_init,
-        eta_init,
-        spin1z_init,
-        spin2z_init,
-    ])[numpy.newaxis]
-else:
-    initial_point = None
-
 with scheme_context:
     logging.info('Starting optimization')
 
     optimize_func = optimize_funcs[args.optimizer]
-    opt_params = optimize_func(bounds, args, extra_args, initial_point)
+    opt_params = optimize_func(bounds, args, extra_args)
 
     logging.info('Optimization complete')