def neuron_activation(u, previous_state, Vin, Wres):
"""
Get reservoir response for a signal.
:param u: input signal. Scalar or vector.
:param previous_state: previous reservoir state.
:param Vin: matrix of input reservoir weights.
:param Wres: matrix of reservoir weights.
:return: reservoir response.
"""
input_activation = Vin.dot(u)
assert input_activation.shape == Vin.shape, 'input activation wrong shape'
recurrent_activation = previous_state.dot(Wres) # activation from neurons
X = sigmoid_af(input_activation + recurrent_activation) # K x N
return X
def harvest_state(data, Vin, Wres, N, P):
"""
Harvest reservoir states for input data.
"""
X = np.zeros((P + 1, N)) # reservoir node's states per input
# Harvest reservoir states
for i in xrange(1, P):
for j in xrange(N):
input_activation = data[0][i - 1].dot(
Vin[:, j]) # activation from input
recurrent_activation = X[i - 1].dot(
Wres[:, j]) # activation from neurons
X[i][j] = sigmoid_af(input_activation + recurrent_activation)
X = np.delete(X, 0, axis=0) # delete zero state initial vector; P x N
return X
def __harvest_states(self, data, P, label):
"""
Harvest reservoir states for input data.
"""
X = np.zeros((P + 1, self.N)) # reservoir node's states per input
label = '{0: <13}'.format(label)
# Harvest reservoir states
bar = pb.ProgressBar(
maxval=P,
widgets=[label, pb.Bar('=', '[', ']'), ' ',
pb.Percentage()]).start()
for i in xrange(1, P):
for j in xrange(self.N):
input_activation = data[0][i - 1].dot(
self.Vin[:, j]) # activation from input
recurrent_activation = X[i - 1].dot(
self.Wres[:, j]) # activation from neurons
X[i][j] = sigmoid_af(input_activation + recurrent_activation)
bar.update(i)
bar.finish()
X = np.delete(X, 0, axis=0) # delete zero state initial vector; P x N
return X
def __harvest_state(u, previous_state, Vin, Wres):
input_activation = Vin.dot(u) # activation from input
recurrent_activation = previous_state.dot(Wres) # activation from neurons
X = sigmoid_af(input_activation + recurrent_activation)
return X
def __harvest_state(self, u, previous_state):
input_activation = self.Vin.dot(u) # activation from input
recurrent_activation = previous_state.dot(
self.Wres) # activation from neurons
X = sigmoid_af(input_activation + recurrent_activation)
return X