def worker(experimental_parameters):
random.seed()
np.random.seed()
results = []
for parameter_set in experimental_parameters:
reservoir = graphgen.uniform_weighted_directed_lfr_graph_asarray(
N=parameter_set['N'],
mu=parameter_set['mu'],
k=parameter_set['k'],
maxk=parameter_set['maxk'],
minc=parameter_set['minc'],
maxc=parameter_set['maxc'],
deg_exp=parameter_set['deg_exp'],
temp_dir_ID=parameter_set['temp_dir_ID'],
full_path=parameter_set['full_path'],
weight_bounds=parameter_set['reservoir_weight_scale'] *
np.array(parameter_set['reservoir_weight_bounds']))
esn = echostatenetwork.DESN(reservoir,
neuron_type=parameter_set['neuron_type'],
neuron_pars=parameter_set['neuron_pars'],
init_state=parameter_set['ic'])
results.append(
(parameter_set[parameter_set['q1']],
parameter_set[parameter_set['q2']],
fixed_point_estimator.fixed_point_estimate(
esn, parameter_set['num_trials'], parameter_set['tmax'],
parameter_set['distance_thresh'], parameter_set['epsilon'],
parameter_set['convergence_delay'])))
return results
def generate_esn(self):
"""
"""
self.reservoir = self.generator_reservoir()
self.input_weights = self.generate_input_weights()
return echostatenetwork.DESN(self.reservoir, self.input_weights,
neuron_type=self.neuron_type, output_type=self.output_neuron_type,
neuron_pars=self.neuron_pars, output_neuron_pars=self.output_neuron_pars,
init_state=self.init_state)
def __init__(self,
network,
input_weight_scale,
signal_ratio,
niter,
input_series,
neuron_type="tanh",
initialization_state="zeros",
neuron_pars=None,
input_weight_bounds=(-1.0, 1.0),
community_key=None,
seed=1,
target_community=None):
"""
input_series needs to be a Lx1x1 vector or a list of Kx1 vectors. L is the length of the input. Simulation time exceeding L will assume inputs of 0
"""
self.network = network
self.signal_ratio = signal_ratio
self.input_weight_scale = input_weight_scale
self.state_history = []
self.initialization_state = initialization_state
self.niter = niter
self.default_input = np.zeros((1, 1))
self.neuron_type = neuron_type
self.neuron_pars = neuron_pars
self.input_series = input_series
self.input_weight_bounds = input_weight_bounds
self.community_key = community_key
self.target_community = target_community
# Generate input weights
self.input_weights = self.CreateInputWeights(
target_community=self.target_community,
community_key=self.community_key)
self.reservoir = nx.adjacency_matrix(
self.network
).A # toarray() may need to be added for different versions of networkx
self.esn_model = echostatenetwork.DESN(
self.reservoir,
self.input_weights,
self.neuron_type,
init_state=self.initialization_state,
neuron_pars=self.neuron_pars)
def generate_esn(self):
"""
"""
self.reservoir = self.generator_reservoir()
# Adjust spectral radius if not None
if self.target_spectral_radius != None:
spectral_radius = np.sort(
np.absolute(np.linalg.eigvals(self.reservoir)))[-1]
self.reservoir = self.reservoir / spectral_radius * self.target_spectral_radius
self.input_weights = self.generate_input_weights()
return echostatenetwork.DESN(
self.reservoir,
self.input_weights,
neuron_type=self.neuron_type,
output_type=self.output_neuron_type,
neuron_pars=self.neuron_pars,
output_neuron_pars=self.output_neuron_pars,
init_state="zeros")
plt.xlim(1, len(activity) + 1)
plt.tight_layout()
plt.savefig(prefix + "_com_activity_ts.png", dpi=300)
plt.clf()
plt.close()
nx_reservoir = graphgen.uniform_weighted_directed_lfr_graph(
N=500,
mu=0.01,
k=6,
maxk=6,
minc=10,
maxc=10,
deg_exp=1.0,
temp_dir_ID=str(np.random.randint(0, 1000)) + 'temptest',
full_path=
"/home/nathaniel/workspace/DiscreteESN/", #/N/u/njrodrig/BigRed2/topology_of_function/
weight_bounds=(-0.2, 1.0))
reservoir = np.asarray(nx.to_numpy_matrix(nx_reservoir))
esn = echostatenetwork.DESN(reservoir,
neuron_type='sigmoid',
neuron_pars={
'c': 1,
'e': 10
},
init_state=(0.0, 0.6))
esn.run_reservoir(100, record=True)
state_history = np.array(esn.network_history)
state_history = state_history.reshape(
(state_history.shape[0], state_history.shape[1]))
plot_community_activity_timeseries("test", nx_reservoir, state_history)