def _preprocess_parameters_for_sampling(self, **parameters):
parameters, number_samples, number_datapoints = broadcast_parent_values(
parameters)
reshaped_parameters, tensor_shape = self._preproces_vector_input(
parameters, self.vector_parameters)
shape = tuple([number_samples, number_datapoints] + tensor_shape)
return reshaped_parameters, shape
def _preprocess_parameters_for_log_prob(self, x, **parameters):
parameters_and_data = parameters
parameters_and_data.update({"x_data": x})
parameters_and_data, number_samples, number_datapoints = broadcast_parent_values(
parameters_and_data)
vector_names = self.vector_parameters
vector_names.add("x_data")
reshaped_parameters_and_data, _ = self._preproces_vector_input(
parameters_and_data, vector_names)
x = reshaped_parameters_and_data.pop("x_data")
return x, reshaped_parameters_and_data, number_samples, number_datapoints
def calculate_log_probability(self, x, z):
"""
One line description
Parameters
----------
Returns
-------
"""
reshaped_dict, n_samples, n_datapoints = broadcast_parent_values({"x": x, "z": z})
labels = np.reshape(reshaped_dict["x"].data, newshape=(n_samples*n_datapoints, 1))
log_probability = -F.softmax_cross_entropy(reshaped_dict["z"], labels, reduce="no")
log_probability = F.reshape(log_probability, shape=(n_samples, n_datapoints))
return log_probability
def _apply_link(
self, parents_values
): #TODO: This is for allowing discrete data, temporary? (for julia) #For Julia: Very important method
cont_values, discrete_values = split_dict(
parents_values, condition=lambda key, val: not is_discrete(val))
if cont_values:
reshaped_dict, number_samples, number_datapoints = broadcast_parent_values(
cont_values)
reshaped_dict.update(discrete_values)
else:
reshaped_dict = discrete_values
reshaped_output = self.link(reshaped_dict)
output = {
key: val.view(size=(number_samples, number_datapoints) +
val.shape[1:]) if is_tensor(val) else val
for key, val in reshaped_output.items()
}
return output
def _apply_link(self, parents_values
): #TODO: This is for allowing discrete data, temporary?
cont_values, discrete_values = split_dict(
parents_values,
condition=lambda key, val: isinstance(val, chainer.Variable))
if cont_values:
reshaped_dict, number_samples, number_datapoints = broadcast_parent_values(
cont_values)
reshaped_dict.update(discrete_values)
else:
reshaped_dict = discrete_values
reshaped_output = self.link(reshaped_dict)
output = {
key: F.reshape(val,
(number_samples, number_datapoints) + val.shape[1:])
if isinstance(val, chainer.Variable) else val
for key, val in reshaped_output.items()
}
return output