def create_thalamic_input(self):
""" This function creates the thalamic neuronal population if this
is specified in stimulus_params.py.
"""
if self.stim_dict['thalamic_input']:
if nest.Rank() == 0:
print('Thalamic input provided')
self.thalamic_population = nest.Create('parrot_neuron',
self.stim_dict['n_thal'])
self.thalamic_weight = get_weight(self.stim_dict['PSP_th'],
self.net_dict)
self.stop_th = (self.stim_dict['th_start'] +
self.stim_dict['th_duration'])
self.poisson_th = nest.Create('poisson_generator')
self.poisson_th.set({
'rate': self.stim_dict['th_rate'],
'start': self.stim_dict['th_start'],
'stop': self.stop_th
})
nest.Connect(self.poisson_th, self.thalamic_population)
self.nr_synapses_th = synapses_th_matrix(self.net_dict,
self.stim_dict)
if self.K_scaling != 1:
self.thalamic_weight = self.thalamic_weight / (self.K_scaling**
0.5)
self.nr_synapses_th = (self.nr_synapses_th * self.K_scaling)
else:
if nest.Rank() == 0:
print('Thalamic input not provided')
def create_thalamic_input(self):
""" This function creates the thalamic neuronal population if this
is specified in stimulus_params.py.
"""
if self.stim_dict['thalamic_input']:
if nest.Rank() == 0:
print('Thalamic input provided')
# Get thalamic weights (using the get_weight() function)
self.thalamic_weight = get_weight(self.stim_dict['PSP_th'],
self.net_dict)
# State when the thalamic input will stop
self.stop_th = (self.stim_dict['th_start'] +
self.stim_dict['th_duration'])
# Define thalamus as a population of parrot neurons
self.thalamic_population = nest.Create('parrot_neuron',
self.stim_dict['n_thal'])
# Create a population of Poisson neurons
self.poisson_th = nest.Create('poisson_generator')
nest.SetStatus(
self.poisson_th, {
'rate': self.stim_dict['th_rate'],
'start': self.stim_dict['th_start'],
'stop': self.stop_th
})
# Connect the Poisson neurons to the thalamus
nest.Connect(self.poisson_th, self.thalamic_population)
# Calculate the number of synapses between thalamus and cortex
self.nr_synapses_th = synapses_th_matrix(self.net_dict,
self.stim_dict)
# Scale the synaptic weights and numbers (if needed)
if self.K_scaling != 1:
self.thalamic_weight = self.thalamic_weight / (self.K_scaling**
0.5)
self.nr_synapses_th = (self.nr_synapses_th * self.K_scaling)
else:
if nest.Rank() == 0:
print('Thalamic input not provided')
def create_thalamic_input(self):
""" This function creates the thalamic neuronal population if this
is specified in stimulus_params.py.
"""
if self.stim_dict['thalamic_input']:
if nest.Rank() == 0:
print('Thalamic input provided')
self.thalamic_population = nest.Create(
'parrot_neuron', self.stim_dict['n_thal']
)
self.thalamic_weight = get_weight(
self.stim_dict['PSP_th'], self.net_dict
)
self.stop_th = (
self.stim_dict['th_start'] + self.stim_dict['th_duration']
)
self.poisson_th = nest.Create('poisson_generator')
nest.SetStatus(
self.poisson_th, {
'rate': self.stim_dict['th_rate'],
'start': self.stim_dict['th_start'],
'stop': self.stop_th
}
)
nest.Connect(self.poisson_th, self.thalamic_population)
self.nr_synapses_th = synapses_th_matrix(
self.net_dict, self.stim_dict
)
if self.K_scaling != 1:
self.thalamic_weight = self.thalamic_weight / (
self.K_scaling ** 0.5)
self.nr_synapses_th = (self.nr_synapses_th * self.K_scaling)
else:
if nest.Rank() == 0:
print('Thalamic input not provided')
def connect_stimulus_vector(self):
""" This function connects the stimulus vector
as specified in stimulus_params.py.
"""
# indices where stimulus vector is not zero
nonzero_stim = np.nonzero(self.stim_dict['PSP_stim'])
PSPs = self.stim_dict['PSP_stim'][nonzero_stim]
stimulus_weight = get_weight(PSPs, self.net_dict)
all_neurons = self.pops[0] + self.pops[1] + self.pops[2]
neurons = [all_neurons[nzs] for nzs in nonzero_stim[0]]
conn_dict = {'rule': 'one_to_one'}
syn_dict = {'weight': stimulus_weight}
rate_vec = np.zeros(np.sum(self.net_dict['N_full']))
if self.stim_dict['stimulus_type'] != 'fixed':
rate_mean = self.stim_dict['stim_rate']
stim_vec = []
for i, stim in enumerate(self.stim_dict['PSP_stim']):
if stim != 0:
rate = np.random.exponential(rate_mean)
rate_vec[i] = rate
poisson_stim = nest.Create('poisson_generator',
params={'rate': rate})
stim_vec.append(poisson_stim[0])
# save stimulus vector
h5.add_to_h5(self.sim_dict['data_path'] + '/results.h5',
{'stim_rate_vec': rate_vec},
'a',
overwrite_dataset=True)
else:
poisson_stim = nest.Create(
'poisson_generator',
params={'rate': self.stim_dict['stim_rate']})
stim_vec = [poisson_stim[0] for i in range(len(neurons))]
nest.Connect(stim_vec, neurons, conn_dict, syn_dict)
def create_populations(self):
""" Creates the neuronal populations.
The neuronal populations are created and the parameters are assigned
to them. The initial membrane potential of the neurons is drawn from a
normal distribution. Scaling of the number of neurons and of the
synapses is performed. If scaling is performed extra DC input is added
to the neuronal populations.
"""
self.N_full = self.net_dict['N_full']
self.N_scaling = self.net_dict['N_scaling']
self.K_scaling = self.net_dict['K_scaling']
self.synapses = get_total_number_of_synapses(self.net_dict)
self.synapses_scaled = self.synapses * self.K_scaling
self.nr_neurons = self.N_full * self.N_scaling
self.K_ext = self.net_dict['K_ext'] * self.K_scaling
self.w_from_PSP = get_weight(self.net_dict['PSP_e'], self.net_dict)
self.weight_mat = get_weight(
self.net_dict['PSP_mean_matrix'], self.net_dict
)
self.weight_mat_std = self.net_dict['PSP_std_matrix']
self.w_ext = self.w_from_PSP
if self.net_dict['poisson_input']:
self.DC_amp_e = np.zeros(len(self.net_dict['populations']))
else:
if nest.Rank() == 0:
print(
'''
no poisson input provided
calculating dc input to compensate
'''
)
self.DC_amp_e = compute_DC(self.net_dict, self.w_ext)
if nest.Rank() == 0:
print(
'The number of neurons is scaled by a factor of: %.2f'
% self.N_scaling
)
print(
'The number of synapses is scaled by a factor of: %.2f'
% self.K_scaling
)
# Scaling of the synapses.
if self.K_scaling != 1:
synapses_indegree = self.synapses / (
self.N_full.reshape(len(self.N_full), 1) * self.N_scaling)
self.weight_mat, self.w_ext, self.DC_amp_e = adj_w_ext_to_K(
synapses_indegree, self.K_scaling, self.weight_mat,
self.w_from_PSP, self.DC_amp_e, self.net_dict, self.stim_dict
)
# Create cortical populations.
self.pops = []
pop_file = open(
os.path.join(self.data_path, 'population_GIDs.dat'), 'w+'
)
for i, pop in enumerate(self.net_dict['populations']):
population = nest.Create(
self.net_dict['neuron_model'], int(self.nr_neurons[i])
)
nest.SetStatus(
population, {
'tau_syn_ex': self.net_dict['neuron_params']['tau_syn_ex'],
'tau_syn_in': self.net_dict['neuron_params']['tau_syn_in'],
'E_L': self.net_dict['neuron_params']['E_L'],
'V_th': self.net_dict['neuron_params']['V_th'],
'V_reset': self.net_dict['neuron_params']['V_reset'],
't_ref': self.net_dict['neuron_params']['t_ref'],
'I_e': self.DC_amp_e[i]
}
)
self.pops.append(population)
pop_file.write('%d %d \n' % (population[0], population[-1]))
pop_file.close()
for thread in np.arange(nest.GetKernelStatus('local_num_threads')):
# Using GetNodes is a work-around until NEST 3.0 is released. It
# will issue a deprecation warning.
local_nodes = nest.GetNodes(
[0], {
'model': self.net_dict['neuron_model'],
'thread': thread
}, local_only=True
)[0]
vp = nest.GetStatus(local_nodes)[0]['vp']
# vp is the same for all local nodes on the same thread
nest.SetStatus(
local_nodes, 'V_m', self.pyrngs[vp].normal(
self.net_dict['neuron_params']['V0_mean'],
self.net_dict['neuron_params']['V0_sd'],
len(local_nodes))
)
def create_populations(self):
""" Creates the neuronal populations.
The neuronal populations are created and the parameters are assigned
to them. The initial membrane potential of the neurons is drawn from a
normal distribution. Scaling of the number of neurons and of the
synapses is performed. If scaling is performed extra DC input is added
to the neuronal populations.
"""
self.N_full = self.net_dict['N_full']
self.N_scaling = self.net_dict['N_scaling']
self.K_scaling = self.net_dict['K_scaling']
self.synapses = get_total_number_of_synapses(self.net_dict)
self.synapses_scaled = self.synapses * self.K_scaling
self.nr_neurons = self.N_full * self.N_scaling
self.K_ext = self.net_dict['K_ext'] * self.K_scaling
self.w_from_PSP = get_weight(self.net_dict['PSP_e'], self.net_dict)
self.weight_mat = get_weight(self.net_dict['PSP_mean_matrix'],
self.net_dict)
self.weight_mat_std = self.net_dict['PSP_std_matrix']
self.w_ext = self.w_from_PSP
if self.net_dict['poisson_input']:
self.DC_amp_e = np.zeros(len(self.net_dict['populations']))
else:
if nest.Rank() == 0:
print("""
no poisson input provided
calculating dc input to compensate
""")
self.DC_amp_e = compute_DC(self.net_dict, self.w_ext)
v0_type_options = ['original', 'optimized']
if self.net_dict['V0_type'] not in v0_type_options:
print('''
'{0}' is not a valid option, replacing it with '{1}'
Valid options are {2}
'''.format(self.net_dict['V0_type'], v0_type_options[0],
v0_type_options))
self.net_dict['V0_type'] = v0_type_options[0]
if nest.Rank() == 0:
print('The number of neurons is scaled by a factor of: %.2f' %
self.N_scaling)
print('The number of synapses is scaled by a factor of: %.2f' %
self.K_scaling)
# Scaling of the synapses.
if self.K_scaling != 1:
synapses_indegree = self.synapses / (
self.N_full.reshape(len(self.N_full), 1) * self.N_scaling)
self.weight_mat, self.w_ext, self.DC_amp_e = adj_w_ext_to_K(
synapses_indegree, self.K_scaling, self.weight_mat,
self.w_from_PSP, self.DC_amp_e, self.net_dict, self.stim_dict)
# Create cortical populations.
self.pops = []
pop_file = open(os.path.join(self.data_path, 'population_nodeids.dat'),
'w+')
for i, pop in enumerate(self.net_dict['populations']):
population = nest.Create(self.net_dict['neuron_model'],
int(self.nr_neurons[i]))
population.set({
'tau_syn_ex':
self.net_dict['neuron_params']['tau_syn_ex'],
'tau_syn_in':
self.net_dict['neuron_params']['tau_syn_in'],
'E_L':
self.net_dict['neuron_params']['E_L'],
'V_th':
self.net_dict['neuron_params']['V_th'],
'V_reset':
self.net_dict['neuron_params']['V_reset'],
't_ref':
self.net_dict['neuron_params']['t_ref'],
'I_e':
self.DC_amp_e[i]
})
if self.net_dict['V0_type'] == 'optimized':
population.set(
V_m=nest.random.normal(mean=self.net_dict['neuron_params']
['V0_mean']['optimized'][i],
std=self.net_dict['neuron_params']
['V0_sd']['optimized'][i]))
elif self.net_dict['V0_type'] == 'original':
population.set(V_m=nest.random.normal(
mean=self.net_dict['neuron_params']['V0_mean']['original'],
std=self.net_dict['neuron_params']['V0_sd']['original']))
self.pops.append(population)
pop_file.write('%d %d \n' % (population[0].get('global_id'),
population[-1].get('global_id')))
pop_file.close()
train_leaky = df_train.loc[:, ['q1_q2_intersect', 'q1_freq', 'q2_freq']]
stops = set(stopwords.words("english"))
df_train['question1'] = df_train['question1'].map(
lambda x: str(x).lower().split())
df_train['question2'] = df_train['question2'].map(
lambda x: str(x).lower().split())
train_qs = pd.Series(df_train['question1'].tolist() +
df_train['question2'].tolist())
words = [x for y in train_qs for x in y]
counts = Counter(words)
weights = {word: get_weight(count) for word, count in counts.items()}
print('\nLoading :: training leaky features...')
df_train_ = pd.read_csv('./data/train_features.csv', encoding="ISO-8859-1")
X_train_ab = df_train_.iloc[:, 2:-1]
X_train_ab = X_train_ab.drop('euclidean_distance', axis=1)
X_train_ab = X_train_ab.drop('jaccard_distance', axis=1)
print('Building :: other training features...')
X_train = build_features(df_train, stops, weights)
X_train = pd.concat((X_train, X_train_ab, train_leaky), axis=1)
y_train = df_train['is_duplicate'].values
try:
print('Saving :: final train features...')
X_train.to_csv('./final_train_features.csv')
def create_populations(self):
""" Creates the neuronal populations.
The neuronal populations are created and the parameters are assigned
to them. The initial membrane potential of the neurons is drawn from a
normal distribution. Scaling of the number of neurons and of the
synapses is performed. If scaling is performed extra DC input is added
to the neuronal populations.
"""
# -------------------------------------------
'''Find model parameters'''
# -------------------------------------------
# Full model parameters
self.N_full = self.net_dict['N_full'] # Full number of neurons
self.synapses = get_total_number_of_synapses(
self.net_dict) # Full number of synapses
# Scaling parameters
self.N_scaling = self.net_dict['N_scaling'] # Neuron number scaling
self.K_scaling = self.net_dict['K_scaling'] # Synapse number scaling
# Scaled parameters
self.nr_neurons = self.N_full * self.N_scaling # Scaled number of neurons
self.synapses_scaled = self.synapses * self.K_scaling # Scaled (internal) synapses
self.K_ext = self.net_dict[
'K_ext'] * self.K_scaling # Scaled (external) synapses
# Calculated weights (to achieve a given change in the membrane potential)
self.w_from_PSP = get_weight(self.net_dict['PSP_e'], self.net_dict)
self.weight_mat = get_weight(self.net_dict['PSP_mean_matrix'],
self.net_dict)
self.weight_mat_std = self.net_dict['PSP_std_matrix']
self.w_ext = self.w_from_PSP
# Network simulated with Poisson input, or DC current?
if self.net_dict['poisson_input']:
self.DC_amp_e = np.zeros(len(self.net_dict['populations']))
else:
if nest.Rank() == 0:
print("""
no poisson input provided
calculating dc input to compensate
""")
self.DC_amp_e = compute_DC(self.net_dict, self.w_ext)
# Scaling of the synapses.
if nest.Rank() == 0:
print('The number of neurons is scaled by a factor of: %.2f' %
self.N_scaling)
print('The number of synapses is scaled by a factor of: %.2f' %
self.K_scaling)
if self.K_scaling != 1:
synapses_indegree = self.synapses / (
self.N_full.reshape(len(self.N_full), 1) * self.N_scaling)
self.weight_mat, self.w_ext, self.DC_amp_e = adj_w_ext_to_K(
synapses_indegree, self.K_scaling, self.weight_mat,
self.w_from_PSP, self.DC_amp_e, self.net_dict, self.stim_dict)
# -------------------------------------------
'''Create cortical population'''
# -------------------------------------------
# Initialise data stores
self.pops = []
pop_file = open(os.path.join(self.data_path, 'population_GIDs.dat'),
'w+')
# Loop over populations
for i, pop in enumerate(self.net_dict['populations']):
# Create a population of a given size
population = nest.Create(self.net_dict['neuron_model'],
int(self.nr_neurons[i]))
# Set the parameters for this population
'''Note that no tau_m is stated here (meaning that all neurons have a default
of 10ms. It is interesting to note that, in the paper by Mejias et al. (2016),
the following tau_m values were used:
For superficial neurons: τE = 6 ms, τI = 15 ms
For infragranular neurons: τE = 30 ms, τI = 75 ms
'''
nest.SetStatus(
population,
{
'tau_syn_ex': self.net_dict['neuron_params']['tau_syn_ex'],
'tau_syn_in': self.net_dict['neuron_params']['tau_syn_in'],
'E_L': self.net_dict['neuron_params']['E_L'],
'V_th': self.net_dict['neuron_params']['V_th'],
'V_reset': self.net_dict['neuron_params']['V_reset'],
't_ref': self.net_dict['neuron_params']['t_ref'],
'I_e': self.DC_amp_e[i] #,
# 'tau_m': self.net_dict['time_constants'][i]
})
# Save population to file
self.pops.append(population)
pop_file.write('%d %d \n' % (population[0], population[-1]))
pop_file.close()
# Something about processing threads
for thread in np.arange(nest.GetKernelStatus('local_num_threads')):
# Using GetNodes is a work-around until NEST 3.0 is released. It
# will issue a deprecation warning.
local_nodes = nest.GetNodes([0], {
'model': self.net_dict['neuron_model'],
'thread': thread
},
local_only=True)[0]
vp = nest.GetStatus(local_nodes)[0]['vp']
# vp is the same for all local nodes on the same thread
nest.SetStatus(
local_nodes, 'V_m', self.pyrngs[vp].normal(
self.net_dict['neuron_params']['V0_mean'],
self.net_dict['neuron_params']['V0_sd'], len(local_nodes)))