def default_q(self):
if "A" in self.inputs_random:
q_A = self.inputs_random["A"].q_distribution()
else:
q_A = self.inputs_nonrandom["A"]
std = positive_exp(shape=self.shape)
return NoisyCumulativeSum(A=q_A,
std=std,
shape=self.shape,
name="q_" + self.name)
def default_q(self):
if "weights" in self.inputs_random:
q_weights = self.inputs_random["weights"].q_distribution()
else:
q_weights = self.inputs_nonrandom["weights"]
if "centers" in self.inputs_random:
q_centers = self.inputs_random["centers"].q_distribution()
else:
q_centers = self.inputs_nonrandom["centers"]
std = positive_exp(shape=self.shape)
return GMMClustering(weights=q_weights,
centers=q_centers,
std=std,
shape=self.shape,
name="q_" + self.name)
def default_q(self):
if "A" in self.inputs_random:
q_A = self.inputs_random["A"].q_distribution()
else:
q_A = self.inputs_nonrandom["A"]
if "B" in self.inputs_random:
q_B = self.inputs_random["B"].q_distribution()
else:
q_B = self.inputs_nonrandom["B"]
std = positive_exp(shape=self.shape)
return NoisyGaussianMatrixProduct(A=q_A,
B=q_B,
std=std,
rescale=self.rescale,
shape=self.shape,
name="q_" + self.name)
def inputs(self):
pe = lambda shape=None, name=None: positive_exp(
shape=shape, name=name, init_log=np.float32(np.ones(shape) * -2.))
return {"mean": unconstrained, "std": pe}