def get_mpe_path(self, root):
"""Assemble TF operations computing the true branch counts for the MPE
downward path through the SPN rooted in ``root``.
Args:
root (Node): The root node of the SPN graph.
"""
def down_fun(node, parent_vals):
self._true_counts[node] = summed = self._accumulate_parents(*parent_vals)
basesum_kwargs = dict(
add_random=self._add_random, use_unweighted=self._use_unweighted,
sample=self._sample, sample_prob=self._sample_prob)
if node.is_op:
kwargs = basesum_kwargs if isinstance(node, BaseSum) else dict()
# Compute for inputs
with tf.name_scope(node.name):
return node._compute_log_mpe_path(
summed, *[self._value.values[i.node] if i else None
for i in node.inputs], **kwargs)
# Generate values if not yet generated
if not self._value.values:
self._value.get_value(root)
with tf.name_scope("TrueMPEPath"):
# Compute the tensor to feed to the root node
graph_input = self._graph_input(self._value.values[root])
# Traverse the graph computing counts
self._true_counts = {}
compute_graph_up_down(root, down_fun=down_fun, graph_input=graph_input)
def get_gradients(self, root):
"""Assemble TF operations computing the gradients of the SPN rooted in
``root``.
Args:
root (Node): The root node of the SPN graph.
"""
def down_fun(node, parent_vals):
# Sum up all parent vals
parent_vals = [pv for pv in parent_vals if pv is not None]
if len(parent_vals) > 1:
summed = tf.add_n(parent_vals, name=node.name + "_add")
else:
summed = parent_vals[0]
self._true_gradients[node] = summed
if node.is_op:
# Compute for inputs
if isinstance(node, BaseSum):
kwargs = dict(
dropconnect_keep_prob=self._dropconnect_keep_prob)
else:
kwargs = dict()
with tf.name_scope(node.name):
if self._log:
return node._compute_log_gradient(
summed, *[
self._value.values[i.node] if i else None
for i in node.inputs
], **kwargs)
else:
return node._compute_log_gradient(
summed, *[
self._value.values[i.node] if i else None
for i in node.inputs
], **kwargs)
# Generate values if not yet generated
if not self._value.values:
self._value.get_value(root)
with tf.name_scope("Gradient"):
# Compute the tensor to feed to the root node
graph_input = tf.ones_like(self._value.values[root])
# Traverse the graph computing gradients
self._true_gradients = {}
compute_graph_up_down(root,
down_fun=down_fun,
graph_input=graph_input)
def get_mpe_path(self, root):
"""Assemble TF operations computing the branch counts for the MPE
downward path through the SPN rooted in ``root``.
Args:
root (Node): The root node of the SPN graph.
"""
def down_fun(node, parent_vals):
# Sum up all parent vals
if len(parent_vals) > 1:
summed = tf.add_n(parent_vals, name=node.name + "_add")
else:
summed = parent_vals[0]
self._counts[node] = summed
if node.is_op:
# Compute for inputs
with tf.name_scope(node.name):
if self._log:
return node._compute_log_mpe_path(
summed,
*[
self._value.values[i.node] if i else None
for i in node.inputs
],
add_random=self._add_random,
use_unweighted=self._use_unweighted)
else:
return node._compute_mpe_path(
summed,
*[
self._value.values[i.node] if i else None
for i in node.inputs
],
add_random=self._add_random,
use_unweighted=self._use_unweighted)
# Generate values if not yet generated
if not self._value.values:
self._value.get_value(root)
with tf.name_scope("MPEPath"):
# Compute the tensor to feed to the root node
graph_input = tf.ones_like(self._value.values[root])
# Traverse the graph computing counts
self._counts = {}
compute_graph_up_down(root,
down_fun=down_fun,
graph_input=graph_input)