def run_one_optimisation(ref, seed, population_size, max_evaluations,
num_selected, num_offspring, mutation_rate,
num_elites, simulator, nogui):
run_optimisation(prefix=ref,
neuroml_file='models/RS/SSTest.net.nml',
target='network_RS',
parameters=parameters,
max_constraints=max_constraints,
min_constraints=min_constraints,
weights=weights,
target_data=target_data,
sim_time=sim_time,
dt=0.025,
seed=seed,
population_size=population_size,
max_evaluations=max_evaluations,
num_selected=num_selected,
num_offspring=num_offspring,
mutation_rate=mutation_rate,
num_elites=num_elites,
simulator=simulator,
nogui=nogui)
def run_one_optimisation(ref,
seed,
population_size,
max_evaluations,
num_selected,
num_offspring,
mutation_rate,
num_elites,
simulator,
nogui,
parameters,
max_constraints,
min_constraints,
neuroml_file = 'prototypes/RS/AllenTest.net.nml',
target = 'network_RS',
weights = weights,
target_data = target_data,
dt = 0.025):
ref = '%s__s%s_p%s_m%s_s%s_o%s_m%s_e%s'%(ref,
seed,
population_size,
max_evaluations,
num_selected,
num_offspring,
mutation_rate,
num_elites)
return run_optimisation(prefix = ref,
neuroml_file = neuroml_file,
target = target,
parameters = parameters,
max_constraints = max_constraints,
min_constraints = min_constraints,
weights = weights,
target_data = target_data,
sim_time = 1500,
dt = dt,
seed = seed,
population_size = population_size,
max_evaluations = max_evaluations,
num_selected = num_selected,
num_offspring = num_offspring,
mutation_rate = mutation_rate,
num_elites = num_elites,
simulator = simulator,
nogui = nogui)
def run_one_optimisation(ref,
seed,
population_size,
max_evaluations,
num_selected,
num_offspring,
mutation_rate,
num_elites,
simulator,
nogui,
parameters,
max_constraints,
min_constraints,
neuroml_file='prototypes/RS/AllenTest.net.nml',
target='network_RS',
weights=weights,
target_data=target_data,
dt=0.025):
ref = '%s__s%s_p%s_m%s_s%s_o%s_m%s_e%s' % (
ref, seed, population_size, max_evaluations, num_selected,
num_offspring, mutation_rate, num_elites)
return run_optimisation(prefix=ref,
neuroml_file=neuroml_file,
target=target,
parameters=parameters,
max_constraints=max_constraints,
min_constraints=min_constraints,
weights=weights,
target_data=target_data,
sim_time=1500,
dt=dt,
seed=seed,
population_size=population_size,
max_evaluations=max_evaluations,
num_selected=num_selected,
num_offspring=num_offspring,
mutation_rate=mutation_rate,
num_elites=num_elites,
simulator=simulator,
nogui=nogui)
target_data = {
max_peak_no: 34,
average_maximum: 30.72,
average_minimum: -75
}
simulator = 'jNeuroML_NEURON'
simulator = 'jNeuroML'
run_optimisation(prefix='TestHHpy',
neuroml_file='test_data/HHCellNetwork.net.nml',
target='HHCellNetwork',
parameters=parameters,
max_constraints=max_constraints,
min_constraints=min_constraints,
weights=weights,
target_data=target_data,
sim_time=700,
population_size=30,
max_evaluations=100,
num_selected=10,
num_offspring=10,
mutation_rate=0.15,
num_elites=1,
seed=12345,
simulator=simulator,
nogui=nogui,
known_target_values=known_target_values,
num_parallel_evaluations=1)
average_maximum: 22.50,
average_minimum: -62.82}
simulator = 'jNeuroML_NEURON'
run_optimisation(prefix = 'TestPyr',
neuroml_file = neuroml_file,
target = 'network',
parameters = parameters,
max_constraints = max_constraints,
min_constraints = min_constraints,
weights = weights,
target_data = target_data,
sim_time = sim_time,
population_size = 20,
max_evaluations = 100,
num_selected = 10,
num_offspring = 10,
mutation_rate = 0.5,
num_elites = 3,
seed = 12345,
simulator = simulator,
nogui = nogui,
known_target_values = known_target_values)
else:
simulator = 'jNeuroML_NEURON'
target_data = {mean_spike_frequency: 32.49, average_maximum: 22.50, average_minimum: -62.82}
simulator = "jNeuroML_NEURON"
run_optimisation(
prefix="TestPyr",
neuroml_file=neuroml_file,
target="network",
parameters=parameters,
max_constraints=max_constraints,
min_constraints=min_constraints,
weights=weights,
target_data=target_data,
sim_time=sim_time,
population_size=20,
max_evaluations=100,
num_selected=10,
num_offspring=10,
mutation_rate=0.5,
num_elites=3,
seed=12345,
simulator=simulator,
nogui=nogui,
known_target_values=known_target_values,
)
else:
simulator = "jNeuroML_NEURON"
target_data = {max_peak_no: 34,
average_maximum: 30.72,
average_minimum: -75}
simulator = 'jNeuroML_NEURON'
simulator = 'jNeuroML'
run_optimisation(prefix = 'TestHHpy',
neuroml_file = 'test_data/HHCellNetwork.net.nml',
target = 'HHCellNetwork',
parameters = parameters,
max_constraints = max_constraints,
min_constraints = min_constraints,
weights = weights,
target_data = target_data,
sim_time = 700,
population_size = 20,
max_evaluations = 60,
num_selected = 10,
num_offspring = 10,
mutation_rate = 0.5,
num_elites = 3,
seed = 12345,
simulator = simulator,
nogui = nogui,
known_target_values = known_target_values,
num_parallel_evaluations = 10)
'cell:Granule_98/channelDensity:GranPassiveCond_all/mS_per_cm2':
0.0330033,
'cell:Granule_98/specificCapacitance:all/uF_per_cm2': 1
}
report = run_optimisation(prefix='Pas_' + prefix,
neuroml_file=neuroml_file,
target=target,
parameters=parameters,
max_constraints=max_constraints_pas,
min_constraints=min_constraints_pas,
weights=weights_pas,
target_data=target_data_pas,
sim_time=sim_time,
dt=dt,
population_size=population_size,
max_evaluations=max_evaluations,
num_selected=num_selected,
num_offspring=num_offspring,
mutation_rate=mutation_rate,
num_elites=num_elites,
simulator=simulator,
nogui=nogui,
show_plot_already=False,
seed=seed,
known_target_values=known_target_values)
plot_baseline_data()
eps = 0.01
'cell:Granule_98/channelDensity:Gran_KDr_98_all/mS_per_cm2':8.89691,
'cell:Granule_98/channelDensity:Gran_H_98_all/mS_per_cm2': 0.03090506,
'cell:Granule_98/channelDensity:GranPassiveCond_all/mS_per_cm2': 0.0330033,
'cell:Granule_98/specificCapacitance:all/uF_per_cm2': 1}
report = run_optimisation(prefix = 'Pas_'+prefix,
neuroml_file = neuroml_file,
target = target,
parameters = parameters,
max_constraints = max_constraints_pas,
min_constraints = min_constraints_pas,
weights = weights_pas,
target_data = target_data_pas,
sim_time = sim_time,
dt = dt,
population_size = population_size,
max_evaluations = max_evaluations,
num_selected = num_selected,
num_offspring = num_offspring,
mutation_rate = mutation_rate,
num_elites = num_elites,
simulator = simulator,
nogui = nogui,
show_plot_already = False,
seed = seed,
known_target_values = known_target_values)
plot_baseline_data()
eps = 0.01
def tune_izh_model(acq_list: List, metrics_from_data: Dict,
currents: Dict) -> Dict:
"""Tune networks model against the data.
Here we generate a network with the necessary number of Izhikevich cells,
one for each current stimulus, and tune them against the experimental data.
:param acq_list: list of indices of acquisitions/sweeps to tune against
:type acq_list: list
:param metrics_from_data: dictionary with the sweep number as index, and
the dictionary containing metrics generated from the analysis
:type metrics_from_data: dict
:param currents: dictionary with sweep number as index and stimulus current
value
"""
# length of simulation of the cells---should match the length of the
# experiment
sim_time = 1500.0
# Create a NeuroML template network simulation file that we will use for
# the tuning
template_doc = NeuroMLDocument(id="IzhTuneNet")
# Add an Izhikevich cell with some parameters to the document
template_doc.izhikevich2007_cells.append(
Izhikevich2007Cell(
id="Izh2007",
C="100pF",
v0="-60mV",
k="0.7nS_per_mV",
vr="-60mV",
vt="-40mV",
vpeak="35mV",
a="0.03per_ms",
b="-2nS",
c="-50.0mV",
d="100pA",
))
template_doc.networks.append(Network(id="Network0"))
# Add a cell for each acquisition list
popsize = len(acq_list)
template_doc.networks[0].populations.append(
Population(id="Pop0", component="Izh2007", size=popsize))
# Add a current source for each cell, matching the currents that
# were used in the experimental study.
counter = 0
for acq in acq_list:
template_doc.pulse_generators.append(
PulseGenerator(
id="Stim{}".format(counter),
delay="80ms",
duration="1000ms",
amplitude="{}pA".format(currents[acq]),
))
template_doc.networks[0].explicit_inputs.append(
ExplicitInput(target="Pop0[{}]".format(counter),
input="Stim{}".format(counter)))
counter = counter + 1
# Print a summary
print(template_doc.summary())
# Write to a neuroml file and validate it.
reference = "TuneIzhFergusonPyr3"
template_filename = "{}.net.nml".format(reference)
write_neuroml2_file(template_doc, template_filename, validate=True)
# Now for the tuning bits
# format is type:id/variable:id/units
# supported types: cell/channel/izhikevich2007cell
# supported variables:
# - channel: vShift
# - cell: channelDensity, vShift_channelDensity, channelDensityNernst,
# erev_id, erev_ion, specificCapacitance, resistivity
# - izhikevich2007Cell: all available attributes
# we want to tune these parameters within these ranges
# param: (min, max)
parameters = {
"izhikevich2007Cell:Izh2007/C/pF": (100, 300),
"izhikevich2007Cell:Izh2007/k/nS_per_mV": (0.01, 2),
"izhikevich2007Cell:Izh2007/vr/mV": (-70, -50),
"izhikevich2007Cell:Izh2007/vt/mV": (-60, 0),
"izhikevich2007Cell:Izh2007/vpeak/mV": (35, 70),
"izhikevich2007Cell:Izh2007/a/per_ms": (0.001, 0.4),
"izhikevich2007Cell:Izh2007/b/nS": (-10, 10),
"izhikevich2007Cell:Izh2007/c/mV": (-65, -10),
"izhikevich2007Cell:Izh2007/d/pA": (50, 500),
} # type: Dict[str, Tuple[float, float]]
# Set up our target data and so on
ctr = 0
target_data = {}
weights = {}
for acq in acq_list:
# data to fit to:
# format: path/to/variable:metric
# metric from pyelectro, for example:
# https://pyelectro.readthedocs.io/en/latest/pyelectro.html?highlight=mean_spike_frequency#pyelectro.analysis.mean_spike_frequency
mean_spike_frequency = "Pop0[{}]/v:mean_spike_frequency".format(ctr)
average_last_1percent = "Pop0[{}]/v:average_last_1percent".format(ctr)
first_spike_time = "Pop0[{}]/v:first_spike_time".format(ctr)
# each metric can have an associated weight
weights[mean_spike_frequency] = 1
weights[average_last_1percent] = 1
weights[first_spike_time] = 1
# value of the target data from our data set
target_data[mean_spike_frequency] = metrics_from_data[acq][
"{}:mean_spike_frequency".format(acq)]
target_data[average_last_1percent] = metrics_from_data[acq][
"{}:average_last_1percent".format(acq)]
target_data[first_spike_time] = metrics_from_data[acq][
"{}:first_spike_time".format(acq)]
# only add these if the experimental data includes them
# these are only generated for traces with spikes
if "{}:average_maximum".format(acq) in metrics_from_data[acq]:
average_maximum = "Pop0[{}]/v:average_maximum".format(ctr)
weights[average_maximum] = 1
target_data[average_maximum] = metrics_from_data[acq][
"{}:average_maximum".format(acq)]
if "{}:average_minimum".format(acq) in metrics_from_data[acq]:
average_minimum = "Pop0[{}]/v:average_minimum".format(ctr)
weights[average_minimum] = 1
target_data[average_minimum] = metrics_from_data[acq][
"{}:average_minimum".format(acq)]
ctr = ctr + 1
# simulator to use
simulator = "jNeuroML"
return run_optimisation(
# Prefix for new files
prefix="TuneIzh",
# Name of the NeuroML template file
neuroml_file=template_filename,
# Name of the network
target="Network0",
# Parameters to be fitted
parameters=list(parameters.keys()),
# Our max and min constraints
min_constraints=[v[0] for v in parameters.values()],
max_constraints=[v[1] for v in parameters.values()],
# Weights we set for parameters
weights=weights,
# The experimental metrics to fit to
target_data=target_data,
# Simulation time
sim_time=sim_time,
# EC parameters
population_size=100,
max_evaluations=500,
num_selected=30,
num_offspring=50,
mutation_rate=0.9,
num_elites=3,
# Seed value
seed=12345,
# Simulator
simulator=simulator,
dt=0.025,
show_plot_already='-nogui' not in sys.argv,
save_to_file="fitted_izhikevich_fitness.png",
save_to_file_scatter="fitted_izhikevich_scatter.png",
save_to_file_hist="fitted_izhikevich_hist.png",
save_to_file_output="fitted_izhikevich_output.png",
num_parallel_evaluations=4,
)
