def compute_output(self, input_data, do_sampling=True):
"""
Compute hidden state
:param input_data: numpy ndarray
:param do_sampling: do binary sample or not
:return: data representation in hidden space
"""
if len(input_data.shape) == 1:
input_data.shape = (1, input_data.shape[0])
h = sigmoid(np.dot(np.c_[np.ones(input_data.shape[0]), input_data], np.vstack((self.b, self.W))))
if do_sampling:
h = self.sample(h)
return h
def compute_output(self, input_data, do_sampling=True):
"""
Compute hidden state
:param input_data: numpy ndarray
:param do_sampling: do binary sample or not
:return: data representation in hidden space
"""
if len(input_data.shape) == 1:
input_data.shape = (1, input_data.shape[0])
h = sigmoid(
np.dot(np.c_[np.ones(input_data.shape[0]), input_data],
np.vstack((self.b, self.W))))
if do_sampling:
h = self.sample(h)
return h
def generate_input(self, input_data, do_sampling=False, pois_N=None):
"""
Restore input data using hidden space representation
:param input_data: data representation in hidden space
:param do_sampling: do_sampling: do binary sample or not (doesn't matter in gaus-bin mode)
:return: data representation in original space
"""
if self.mode == 'gaus-bin':
return np.dot(np.c_[np.ones(input_data.shape[0]), input_data], np.vstack((self.a, self.W.T)))
elif self.mode == 'pois-bin':
e = np.exp(np.dot(np.c_[np.ones(input_data.shape[0]), input_data], np.vstack((self.a, self.W.T))))
if pois_N is None:
return (e.T/np.sum(e, axis=1)).T
else:
return (e.T/((1/pois_N) * np.sum(e, axis=1))).T
v = sigmoid(np.dot(np.c_[np.ones(input_data.shape[0]), input_data], np.vstack((self.a, self.W.T))))
if do_sampling:
v = self.sample(v)
return v
def generate_input(self, input_data, do_sampling=False, pois_N=None):
"""
Restore input data using hidden space representation
:param input_data: data representation in hidden space
:param do_sampling: do_sampling: do binary sample or not (doesn't matter in gaus-bin mode)
:return: data representation in original space
"""
if self.mode == 'gaus-bin':
return np.dot(np.c_[np.ones(input_data.shape[0]), input_data],
np.vstack((self.a, self.W.T)))
elif self.mode == 'pois-bin':
e = np.exp(
np.dot(np.c_[np.ones(input_data.shape[0]), input_data],
np.vstack((self.a, self.W.T))))
if pois_N is None:
return (e.T / np.sum(e, axis=1)).T
else:
return (e.T / ((1 / pois_N) * np.sum(e, axis=1))).T
v = sigmoid(
np.dot(np.c_[np.ones(input_data.shape[0]), input_data],
np.vstack((self.a, self.W.T))))
if do_sampling:
v = self.sample(v)
return v
fig, ax = plt.subplots()
ax.plot(t, Ricker, 'k', lw=2)
ax.set(xlabel='time (s)',
ylabel='Amplitude',
title='Function approximation',
ylim=(-0.8, 1.2),
xlim=(-8, 8))
ax.grid()
plt.savefig('Ricker.png', dpi=600)
plt.show()
#===================== activation functions ===============================================
# Sigmoid activation function
sigmo = sigmoid(t)
fig, ax = plt.subplots()
ax.plot(t, sigmo, 'r', lw=1)
ax.set(title='Sigmoid function', ylim=(-0.1, 1.1), xlim=(-8, 8))
ax.grid()
plt.savefig('Sigmoid.png', dpi=600)
plt.show()
# tanh activation function
tan = tanh(t)
fig, ax = plt.subplots()
ax.plot(t, tan, 'r', lw=1)
ax.set(title='Tanh function', ylim=(-1.1, 1.1), xlim=(-8, 8))
fig, ax = plt.subplots()
ax.plot(X, Ricker, 'k', lw=4)
ax.set(xlabel='time (s)',
ylabel='Amplitude',
title='Function approximation',
ylim=(-0.8, 1.2),
xlim=(-8, 8))
ax.grid()
#plt.savefig('Ricker.png',dpi=600)
plt.show()
#===================== activation functions ===============================================
# Sigmoid activation function
sigmo = sigmoid(X)
fig, ax = plt.subplots()
ax.plot(X, sigmo, 'r', lw=4)
ax.set(title='Sigmoid function', ylim=(-0.1, 1.1), xlim=(-8, 8))
ax.grid()
plt.savefig('Sigmoid.png', dpi=600)
plt.show()
# tanh activation function
tan = tanh(X)
fig, ax = plt.subplots()
ax.plot(X, tan, 'r', lw=4)
ax.set(title='Tanh function', ylim=(-1.1, 1.1), xlim=(-8, 8))
