def sample_from_activation(self, vmap):
mu = self.mean_from_activation(vmap)
sigma2 = self.variance_from_activation(vmap)
return samplers.gaussian(mu, sigma2)
def sample_from_activation(self, vmap):
s = vmap[self] + samplers.gaussian(0, T.nnet.sigmoid(vmap[self])) # approximation: linear + gaussian noise
return T.max(0, s) # rectify
def sample_from_activation(self, vmap):
mu = self.mean_from_activation(vmap)
return samplers.gaussian(mu)
def sample_from_activation(self, vmap):
s = vmap[self] + samplers.gaussian(0, T.nnet.sigmoid(
vmap[self])) # approximation: linear + gaussian noise
return T.max(0, s) # rectify
def sample_from_activation(self, vmap):
mu = self.mean_from_activation(vmap)
return samplers.gaussian(mu)
def sample_from_activation(self, vmap):
mu = self.mean_from_activation(vmap)
sigma2 = self.variance_from_activation(vmap)
return samplers.gaussian(mu, sigma2)
def sample_from_activation(self, vmap):
a1 = vmap[self]
a2 = vmap[self.precision_units]
return samplers.gaussian(a1 / (-2 * a2), 1 / (-2 * a2))
def sample_from_activation(self, vmap):
return samplers.gaussian(vmap[self])
def sample_from_activation(self, vmap):
a1 = vmap[self]
a2 = vmap[self.precision_units]
return samplers.gaussian(a1 / (-2 * a2), 1 / (-2 * a2))
def sample_from_activation(self, vmap):
return samplers.gaussian(vmap[self])
