def ga_evaluate(self, min, max, args={}):
"""
Optimization using Genetics algorithm (GA).
:param min: Minimum values for parameters
:param max: Maximum values for parameters
:param args: Optional arguments for neurotune optimizer
:return: optimized values
"""
# Initialize neurotune optional arguments
if 'population_size' in args:
population_size = args['population_size']
else:
population_size = 100
if 'max_evaluations' in args:
max_evaluations = args['max_evaluations']
else:
max_evaluations = 100
if 'num_selected' in args:
num_selected = args['num_selected']
else:
num_selected = None
if 'num_offspring' in args:
num_offspring = args['num_offspring']
else:
num_offspring = None
if 'num_elites' in args:
num_elites = args['num_elites']
else:
num_elites = 1
if 'mutation_rate' in args:
mutation_rate = args['mutation_rate']
else:
mutation_rate = 0.2
if 'maximize' in args:
maximize = args['maximize']
else:
maximize = False
if 'seeds' in args:
seeds = args['seeds']
else:
seeds = []
if 'verbose' in args:
verbose = args['verbose']
else:
verbose = False
candidates = optimizers.CustomOptimizerA(
max_constraints=max,
min_constraints=min,
evaluator=self,
population_size=
population_size, #20 times larger than free parameters
max_evaluations=max_evaluations,
num_selected=num_selected,
num_offspring=num_offspring,
num_elites=num_elites,
mutation_rate=mutation_rate,
maximize=maximize,
seeds=seeds,
verbose=verbose)
best_candidate, score = candidates.optimize(do_plot=False, seed=1234)
return best_candidate, score
def run_optimisation(prefix,
config,
level,
parameters,
max_constraints,
min_constraints,
weights,
target_data,
sim_time=500,
dt=0.05,
analysis_start_time=0,
population_size=20,
max_evaluations=20,
num_selected=10,
num_offspring=20,
mutation_rate=0.9,
num_elites=1,
seed=12345,
simulator='jNeuroML',
nogui=False,
num_local_procesors_to_use=4):
print("Running optimisation...")
print("parameters: %s" % parameters)
print("max_constraints: %s" % max_constraints)
print("min_constraints: %s" % min_constraints)
print("simulator: %s" % simulator)
ref = prefix + config
run_dir = "NT_%s_%s" % (ref, time.ctime().replace(' ', '_').replace(
':', '.'))
os.mkdir(run_dir)
my_controller = C302Controller(
ref,
level,
config,
sim_time,
dt,
simulator=simulator,
generate_dir=run_dir,
num_local_procesors_to_use=num_local_procesors_to_use)
peak_threshold = -31 if level.startswith('A') or level.startswith(
'B') else (-40)
analysis_var = {
'peak_delta': 0,
'baseline': 0,
'dvdt_threshold': 0,
'peak_threshold': peak_threshold
}
data = ref + '.dat'
sim_var = OrderedDict()
for i in range(len(parameters)):
sim_var[
parameters[i]] = max_constraints[i] / 2 + min_constraints[i] / 2
#make an evaluator, using automatic target evaluation:
my_evaluator = c302Evaluators.EnhancedNetworkEvaluator(
controller=my_controller,
analysis_start_time=analysis_start_time,
analysis_end_time=sim_time,
parameters=parameters,
analysis_var=analysis_var,
weights=weights,
targets=target_data)
#make an optimizer
my_optimizer = optimizers.CustomOptimizerA(max_constraints,
min_constraints,
my_evaluator,
population_size=population_size,
max_evaluations=max_evaluations,
num_selected=num_selected,
num_offspring=num_offspring,
num_elites=num_elites,
mutation_rate=mutation_rate,
seeds=None,
verbose=False)
start = time.time()
#run the optimizer
best_candidate, fitness = my_optimizer.optimize(do_plot=False,
seed=seed,
summary_dir=run_dir)
secs = time.time() - start
reportj = {}
info = "Ran %s evaluations (pop: %s) in %f seconds (%f mins total; %fs per eval)\n\n" % (
max_evaluations, population_size, secs, secs / 60.0,
(secs / max_evaluations))
report = "----------------------------------------------------\n\n" + info
reportj['comment'] = info
reportj['time'] = secs
for key, value in zip(parameters, best_candidate):
sim_var[key] = value
best_candidate_t, best_candidate_v = my_controller.run_individual(
sim_var, show=False)
best_candidate_results = my_controller.last_results
best_candidate_analysis = c302Analysis.Data_Analyser(
best_candidate_v,
best_candidate_t,
analysis_var,
start_analysis=analysis_start_time,
end_analysis=sim_time)
best_cand_analysis_full = best_candidate_analysis.analyse()
best_cand_analysis = best_candidate_analysis.analyse(target_data)
report += "---------- Best candidate ------------------------------------------\n"
report += pp.pformat(best_candidate) + "\n\n"
report += pp.pformat(best_cand_analysis_full) + "\n"
report += pp.pformat(best_cand_analysis) + "\n\n"
report += "FITNESS: %f\n\n" % fitness
print(report)
reportj['fitness'] = fitness
reportj['fittest vars'] = dict(sim_var)
reportj['best_cand_details'] = best_candidate
reportj['best_cand_analysis_full'] = best_cand_analysis_full
reportj['best_cand_analysis'] = best_cand_analysis
reportj['parameters'] = parameters
reportj['analysis_var'] = analysis_var
reportj['target_data'] = target_data
reportj['weights'] = weights
reportj['analysis_start_time'] = analysis_start_time
reportj['population_size'] = population_size
reportj['max_evaluations'] = max_evaluations
reportj['num_selected'] = num_selected
reportj['num_offspring'] = num_offspring
reportj['mutation_rate'] = mutation_rate
reportj['num_elites'] = num_elites
reportj['seed'] = seed
reportj['simulator'] = simulator
reportj['sim_time'] = sim_time
reportj['dt'] = dt
reportj['run_directory'] = run_dir
reportj['reference'] = ref
report_file = open("%s/report.json" % run_dir, 'w')
report_file.write(pp.pformat(reportj))
report_file.close()
plot_file = open("%s/plotgens.py" % run_dir, 'w')
plot_file.write(
'from neurotune.utils import plot_generation_evolution\nimport os\n')
plot_file.write('\n')
plot_file.write('parameters = %s\n' % parameters)
plot_file.write('\n')
plot_file.write(
"curr_dir = os.path.dirname(__file__) if len(os.path.dirname(__file__))>0 else '.'\n"
)
plot_file.write(
"plot_generation_evolution(parameters, individuals_file_name = '%s/ga_individuals.csv'%curr_dir)\n"
)
plot_file.close()
if not nogui:
added = []
for wref in weights.keys():
ref = wref.split(':')[0]
if not ref in added and not "phase_offset" in ref:
added.append(ref)
best_candidate_plot = plt.plot(best_candidate_t,
best_candidate_v[ref],
label="%s - %i evaluations" %
(ref, max_evaluations))
plt.legend()
plt.ylim(-80.0, 80.0)
plt.xlim(0.0, sim_time)
plt.title("Models")
plt.xlabel("Time (ms)")
plt.ylabel("Membrane potential(mV)")
utils.plot_generation_evolution(
sim_var.keys(),
individuals_file_name='%s/ga_individuals.csv' % run_dir)
c302_utils.plot_c302_results(best_candidate_results,
config,
level,
directory=run_dir,
save=True)
targets=targets)
population_size = 20
max_evaluations = 300
num_selected = 10
num_offspring = 5
mutation_rate = 0.5
num_elites = 1
#make an optimizer
my_optimizer=optimizers.CustomOptimizerA(max_constraints,
min_constraints,
my_evaluator,
population_size=population_size,
max_evaluations=max_evaluations,
num_selected=num_selected,
num_offspring=num_offspring,
num_elites=num_elites,
mutation_rate=mutation_rate,
seeds=None,
verbose=True)
#run the optimizer
best_candidate, fitness = my_optimizer.optimize(do_plot=False, seed=1234567)
keys = list(sim_vars.keys())
for i in range(len(best_candidate)):
sim_vars[keys[i]] = best_candidate[i]
fit_times, fit_volts = swc.run_individual(sim_vars, showPlots, False)
def main(targets,
population_size=100,
max_evaluations=100,
num_selected=5,
num_offspring=5,
seeds=None):
"""
The optimization runs in this main method
"""
#make a controller
my_controller = BasketCellController()
#parameters to be modified in each simulation
parameters = [
'axon_gbar_na', 'axon_gbar_kv', 'axon_gbar_kv3', 'soma_gbar_na',
'soma_gbar_kv', 'soma_gbar_kv3'
]
#above parameters will not be modified outside these bounds:
min_constraints = [0, 0, 0, 0, 0, 0]
max_constraints = [10000, 30, 1, 300, 20, 2]
# EXAMPLE - how to set a seed
#manual_vals=[50,50,2000,70,70,5,0.1,28.0,49.0,-73.0,23.0]
#analysis variables, these default values will do:
analysis_var = {'peak_delta': 1e-4, 'baseline': 0, 'dvdt_threshold': 0.0}
weights = {
'average_minimum': 1.0,
'spike_frequency_adaptation': 1.0,
'trough_phase_adaptation': 1.0,
'mean_spike_frequency': 1.0,
'average_maximum': 1.0,
'trough_decay_exponent': 1.0,
'interspike_time_covar': 1.0,
'min_peak_no': 1.0,
'spike_broadening': 1.0,
'spike_width_adaptation': 1.0,
'max_peak_no': 1.0,
'first_spike_time': 1.0,
'peak_decay_exponent': 1.0,
'pptd_error': 1.0
}
data = './100pA_1.csv'
print('data location')
print(data)
#make an evaluator, using automatic target evaluation:
my_evaluator = evaluators.IClampEvaluator(controller=my_controller,
analysis_start_time=0,
analysis_end_time=900,
target_data_path=data,
parameters=parameters,
analysis_var=analysis_var,
weights=weights,
targets=targets,
automatic=False)
#make an optimizer
my_optimizer = optimizers.CustomOptimizerA(max_constraints,
min_constraints,
my_evaluator,
population_size=population_size,
max_evaluations=max_evaluations,
num_selected=num_selected,
num_offspring=num_offspring,
num_elites=1,
mutation_rate=0.5,
seeds=seeds,
verbose=True)
#run the optimizer
best_candidate, fitness = my_optimizer.optimize(do_plot=False)
return best_candidate
def _run_optimisation(a):
if isinstance(a.parameters, str):
a.parameters = parse_list_arg(a.parameters)
if isinstance(a.min_constraints, str):
a.min_constraints = parse_list_arg(a.min_constraints)
if isinstance(a.max_constraints, str):
a.max_constraints = parse_list_arg(a.max_constraints)
if isinstance(a.target_data, str):
a.target_data = parse_dict_arg(a.target_data)
if isinstance(a.weights, str): a.weights = parse_dict_arg(a.weights)
if isinstance(a.known_target_values, str):
a.known_target_values = parse_dict_arg(a.known_target_values)
if isinstance(a.extra_report_info, str):
a.extra_report_info = parse_dict_arg(a.extra_report_info)
pynml.print_comment_v(
"====================================================================================="
)
pynml.print_comment_v("Starting run_optimisation with: ")
keys = sorted(a.__dict__.keys())
for key in keys:
value = a.__dict__[key]
pynml.print_comment_v(" %s = %s%s" % (key, ' ' *
(30 - len(key)), value))
pynml.print_comment_v(
"====================================================================================="
)
if a.dry_run:
pynml.print_comment_v("Dry run; not running optimization...")
return
ref = a.prefix
run_dir = "NT_%s_%s" % (ref, time.ctime().replace(' ', '_').replace(
':', '.'))
os.mkdir(run_dir)
my_controller = NeuroMLController(
ref,
a.neuroml_file,
a.target,
a.sim_time,
a.dt,
simulator=a.simulator,
generate_dir=run_dir,
num_parallel_evaluations=a.num_parallel_evaluations,
cleanup=a.cleanup)
peak_threshold = 0
analysis_var = {
'peak_delta': 0,
'baseline': 0,
'dvdt_threshold': 0,
'peak_threshold': peak_threshold
}
sim_var = OrderedDict()
#make an evaluator, using automatic target evaluation:
my_evaluator = evaluators.NetworkEvaluator(
controller=my_controller,
analysis_start_time=a.analysis_start_time,
analysis_end_time=a.sim_time,
parameters=a.parameters,
analysis_var=analysis_var,
weights=a.weights,
targets=a.target_data)
#make an optimizer
my_optimizer = optimizers.CustomOptimizerA(
a.max_constraints,
a.min_constraints,
my_evaluator,
population_size=a.population_size,
max_evaluations=a.max_evaluations,
num_selected=a.num_selected,
num_offspring=a.num_offspring,
num_elites=a.num_elites,
mutation_rate=a.mutation_rate,
seeds=None,
verbose=a.verbose)
start = time.time()
#run the optimizer
best_candidate, fitness = my_optimizer.optimize(do_plot=False,
seed=a.seed,
summary_dir=run_dir)
secs = time.time() - start
reportj = {}
info = "Ran %s evaluations (pop: %s) in %f seconds (%f mins total; %fs per eval)\n\n" % (
a.max_evaluations, a.population_size, secs, secs / 60.0,
(secs / a.max_evaluations))
report = "----------------------------------------------------\n\n" + info
reportj['comment'] = info
reportj['time'] = secs
for key, value in zip(a.parameters, best_candidate):
sim_var[key] = value
best_candidate_t, best_candidate_v = my_controller.run_individual(
sim_var, show=False, cleanup=False)
best_candidate_analysis = analysis.NetworkAnalysis(
best_candidate_v,
best_candidate_t,
analysis_var,
start_analysis=a.analysis_start_time,
end_analysis=a.sim_time)
best_cand_analysis_full = best_candidate_analysis.analyse()
best_cand_analysis = best_candidate_analysis.analyse(a.weights.keys())
report += "---------- Best candidate ------------------------------------------\n"
report += pp.pformat(best_cand_analysis_full) + "\n\n"
report += "TARGETS: \n"
report += pp.pformat(a.target_data) + "\n\n"
report += "TUNED VALUES:\n"
report += pp.pformat(best_cand_analysis) + "\n\n"
report += "FITNESS: %f\n\n" % fitness
report += "FITTEST: %s\n\n" % pp.pformat(dict(sim_var))
pynml.print_comment_v(report)
reportj['fitness'] = fitness
reportj['fittest vars'] = dict(sim_var)
reportj['best_cand_analysis_full'] = best_cand_analysis_full
reportj['best_cand_analysis'] = best_cand_analysis
reportj['parameters'] = a.parameters
reportj['analysis_var'] = analysis_var
reportj['target_data'] = a.target_data
reportj['weights'] = a.weights
reportj['analysis_start_time'] = a.analysis_start_time
reportj['population_size'] = a.population_size
reportj['max_evaluations'] = a.max_evaluations
reportj['num_selected'] = a.num_selected
reportj['num_offspring'] = a.num_offspring
reportj['mutation_rate'] = a.mutation_rate
reportj['num_elites'] = a.num_elites
reportj['seed'] = a.seed
reportj['simulator'] = a.simulator
reportj['sim_time'] = a.sim_time
reportj['dt'] = a.dt
reportj['run_directory'] = run_dir
reportj['reference'] = ref
if a.extra_report_info:
for key in a.extra_report_info:
reportj[key] = a.extra_report_info[key]
report_file = open("%s/report.json" % run_dir, 'w')
report_file.write(pp.pformat(reportj))
report_file.close()
plot_file = open("%s/plotgens.py" % run_dir, 'w')
plot_file.write(
'from neurotune.utils import plot_generation_evolution\nimport os\n')
plot_file.write('\n')
plot_file.write('parameters = %s\n' % a.parameters)
plot_file.write('\n')
plot_file.write(
"curr_dir = os.path.dirname(__file__) if len(os.path.dirname(__file__))>0 else '.'\n"
)
plot_file.write(
"plot_generation_evolution(parameters, individuals_file_name = '%s/ga_individuals.csv'%curr_dir)\n"
)
plot_file.close()
if not a.nogui:
added = []
#print("Plotting saved data from %s which are relevant for targets: %s"%(best_candidate_v.keys(), a.target_data.keys()))
fig = plt.figure()
fig.canvas.set_window_title(
"Simulation of fittest individual from run: %s" % ref)
for tref in best_candidate_v.keys(
): ##################a.target_data.keys():
ref = tref.split(':')[0]
if not ref in added:
added.append(ref)
#pynml.print_comment(" - Adding plot of: %s"%ref)
plt.plot(best_candidate_t,
best_candidate_v[ref],
label="%s - %i evaluations" %
(ref, a.max_evaluations))
plt.legend()
#plt.ylim(-80.0,80.0)
plt.xlim(0.0, a.sim_time)
plt.title("Models %s" % a.prefix)
plt.xlabel("Time (ms)")
plt.ylabel("Membrane potential(mV)")
utils.plot_generation_evolution(
sim_var.keys(),
individuals_file_name='%s/ga_individuals.csv' % run_dir,
target_values=a.known_target_values,
show_plot_already=a.show_plot_already,
title_prefix=ref)
if a.show_plot_already:
plt.show()
return reportj
def main():
"""
The optimization runs in this main method
"""
show_plots = not (len(sys.argv) == 2 and sys.argv[1] == '-nogui')
#make a controller
my_controller= BasketCellController(show_plots)
#parameters to be modified in each simulation
parameters = ['axon_gbar_na',
'axon_gbar_kv',
'axon_gbar_kv3',
'soma_gbar_na',
'soma_gbar_kv',
'soma_gbar_kv3']
#above parameters will not be modified outside these bounds:
min_constraints = [0,0,0,0,0,0]
max_constraints = [10000,30,1,300,20,2]
# EXAMPLE - how to set a seed
#manual_vals=[50,50,2000,70,70,5,0.1,28.0,49.0,-73.0,23.0]
#analysis variables, these default values will do:
analysis_var={'peak_delta':0,
'baseline':0,
'dvdt_threshold':2}
weights={'average_minimum': 1.0,
'spike_frequency_adaptation': 1.0,
'trough_phase_adaptation': 1.0,
'mean_spike_frequency': 1.0,
'average_maximum': 1.0,
'trough_decay_exponent': 1.0,
'interspike_time_covar': 1.0,
'min_peak_no': 1.0,
'spike_broadening': 1.0,
'spike_width_adaptation': 1.0,
'max_peak_no': 1.0,
'first_spike_time': 1.0,
'peak_decay_exponent': 1.0,
'pptd_error':1.0}
#make an evaluator, using automatic target evaluation:
my_evaluator=evaluators.IClampEvaluator(controller=my_controller,
analysis_start_time=1,
analysis_end_time=500,
target_data_path='100pA_1a.csv',
parameters=parameters,
analysis_var=analysis_var,
weights=weights,
targets=None, # because we're using automatic
automatic=True)
#make an optimizer
my_optimizer=optimizers.CustomOptimizerA(max_constraints,min_constraints,my_evaluator,
population_size=3,
max_evaluations=100,
num_selected=3,
num_offspring=3,
num_elites=1,
seeds=None)
#run the optimizer
my_optimizer.optimize(do_plot=show_plots)
# userData['samples'] = {'IClamp':IClamp,'IV':IV}
myInitiator = initiators.Initiator(userData)
sampleData = myInitiator.get_sample_params()
bio_params = myInitiator.get_bio_params()
# sim_params = myInitiator.get_sim_params(type='IClamp')
sim_params = myInitiator.get_sim_params()
myEvaluator = evaluators.Evaluator(sampleData,sim_params,bio_params)
candidates = optimizers.CustomOptimizerA(bio_params['max_val_channel'],
bio_params['min_val_channel'],
myEvaluator,
population_size=300, #20 times larger than free parameters
max_evaluations=600,
num_selected=2,
num_offspring=15,
num_elites=1,
mutation_rate=0.05,
maximize = False,
seeds=None,
verbose=True)
best_candidate = candidates.optimize(do_plot=True, seed=1234)
# best_candidate = [189.66504757196532, 0.04619592278775828, -0.0015715032129984834, 0.03232782689368996,
# 0.0009038799935426481, -0.0006996189007248855, 0.0002076054785033701, 0.5361776032113692,
# 2.0, 1.0, 2.9942088447494227e-07, 0.18712673917281542, -1.1396759086697654e-07,
# 0.014145060464901655, 1.0]
best_candidate_params = dict(zip(bio_params['channel_params'],best_candidate))
cell_var = dict(zip(bio_params['cell_params'],bio_params['val_cell_params']))
mySimulator = simulators.Simulator(sim_params,best_candidate_params,cell_var,bio_params['gate_params'])
bestSim = dict()
