def initialize_weights(self):
# Initialize all weights. biases, and initial values
self.var_dict = {}
# all keys in par with a suffix of '0' are initial values of trainable variables
for k, v in par.items():
if k[-1] == '0':
name = k[:-1]
self.var_dict[name] = tf.Variable(par[k], name)
self.syn_x_init = tf.constant(par['syn_x_init'])
self.syn_u_init = tf.constant(par['syn_u_init'])
if par['EI']:
# ensure excitatory neurons only have postive outgoing weights,
# and inhibitory neurons have negative outgoing weights
self.w_rnn = tf.constant(par['EI_matrix']) @ tf.nn.relu(
self.var_dict['w_rnn'])
else:
self.w_rnn = self.var_dict['w_rnn']
# Make reset operations
reset_vars_ops = []
for k, v in self.var_dict.items():
name = k + "0"
reset_vars_ops.append(tf.assign(v, par[name]))
self.reset_vars_ops = tf.group(*reset_vars_ops)
def initialize_weights(self):
self.var_dict = {}
# all keys in par with a suffix of '0' are initial values of trainable variables
for k, v in par.items():
if k[-1] == '0':
name = k[:-1]
self.var_dict[name] = tf.Variable(par[k], name)
self.syn_x_init = tf.constant(par['syn_x_init'])
self.syn_u_init = tf.constant(par['syn_u_init'])
if par['EI']:
# ensure excitatory neurons only have postive outgoing weights,
# and inhibitory neurons have negative outgoing weights
self.w_rnn = tf.constant(par['EI_matrix']) @ tf.nn.relu(
self.var_dict['w_rnn'])
else:
self.w_rnn = self.var_dict['w_rnn']