def __generate_information_probability(
self,
related_node_target_links
):
"""Calculate probabilities with target links.
It creates the entropy and probabilities from related nodes in
whole AtomSpace.
Args:
related_node_target_links: Target link tuples in related node list
:param related_node_target_links: list[list[EqualLinkKey]]
Returns:
The max value of n_gram.
:rtype : int
"""
log.debug(
"ConnectConflictInteractionInformation: Calculating probabilities "
"(Total: " + str(len(related_node_target_links)) + ")"
)
current_ratio = 0
# Register the every link in related nodes to provider.
for i, target_equal_link in enumerate(related_node_target_links):
current_ratio = self.__print_progress(
"ConnectConflictInteractionInformation:PROB:",
current_ratio, i, len(related_node_target_links), 30
)
# TODO: To prevent freeze during probability generation,
# user can limit the max value of calculation.
max_repeat_length = self.provider.n_gram \
if 0 < self.provider.n_gram < len(target_equal_link) \
else len(target_equal_link)
# Make n-gram data in each related node.
# The provider with high n-gram will provides more correct data,
# but speed will going slower rapidly.
# (0, 0, 0), (0, 0, 1), (0, 1, 0), (0, 1, 1), ... (1, 1, 1)
cartesian_binary_iterator = \
itertools.product([False, True], repeat=max_repeat_length)
for viable_case_binary in enumerate(cartesian_binary_iterator):
gram_data = list()
for j, selector in enumerate(viable_case_binary):
# Make each gram data.
if selector:
gram_data.append(eq_link.key_to_link(
self.a,
target_equal_link[j],
self.ret[0],
target_equal_link[j].tv
))
# Register the generated gram_data.
self.provider.add_one_rawdata_count(
[data for data in gram_data if data is not None], 1
)
# Update provider's statistic data.
PyProbabilityAtom().calculate_probabilities(self.provider)
PyEntropyAtom().calculate_entropies(self.provider)
def __print_progress(
self, msg, current_ratio, current_count, total, step=10
):
# To print(debug) progress of evaluating.
if current_ratio < current_count:
current_ratio += total * step * 0.01
log.debug(
msg + ": " + str(current_count) + "/" + str(total) +
" (" + str(100 * current_count / float(total)) + "%)"
)
return current_ratio
def __print_progress(self,
msg,
current_ratio,
current_count,
total,
step=10):
# To print(debug) progress of evaluating.
if current_ratio < current_count:
current_ratio += total * step * 0.01
log.debug(msg + ": " + str(current_count) + "/" + str(total) +
" (" + str(100 * current_count / float(total)) + "%)")
return current_ratio
def __generate_information_probability(self, related_node_target_links):
"""Calculate probabilities with target links.
It creates the entropy and probabilities from related nodes in
whole AtomSpace.
Args:
related_node_target_links: Target link tuples in related node list
:param related_node_target_links: list[list[EqualLinkKey]]
Returns:
The max value of n_gram.
:rtype : int
"""
log.debug(
"ConnectConflictInteractionInformation: Calculating probabilities "
"(Total: " + str(len(related_node_target_links)) + ")")
current_ratio = 0
# Register the every link in related nodes to provider.
for i, target_equal_link in enumerate(related_node_target_links):
current_ratio = self.__print_progress(
"ConnectConflictInteractionInformation:PROB:", current_ratio,
i, len(related_node_target_links), 30)
# TODO: To prevent freeze during probability generation,
# user can limit the max value of calculation.
max_repeat_length = self.provider.n_gram \
if 0 < self.provider.n_gram < len(target_equal_link) \
else len(target_equal_link)
# Make n-gram data in each related node.
# The provider with high n-gram will provides more correct data,
# but speed will going slower rapidly.
# (0, 0, 0), (0, 0, 1), (0, 1, 0), (0, 1, 1), ... (1, 1, 1)
cartesian_binary_iterator = \
itertools.product([False, True], repeat=max_repeat_length)
for viable_case_binary in enumerate(cartesian_binary_iterator):
gram_data = list()
for j, selector in enumerate(viable_case_binary):
# Make each gram data.
if selector:
gram_data.append(
eq_link.key_to_link(self.a, target_equal_link[j],
self.ret[0],
target_equal_link[j].tv))
# Register the generated gram_data.
self.provider.add_one_rawdata_count(
[data for data in gram_data if data is not None], 1)
# Update provider's statistic data.
PyProbabilityAtom().calculate_probabilities(self.provider)
PyEntropyAtom().calculate_entropies(self.provider)
def __make_result_log(self, result_list, reverse):
# To print(debug) interaction information value.
if reverse:
result_list = reversed(result_list)
for i, result in enumerate(result_list):
name = ""
# Prints only top 5 results.
if i < 15:
for link in result['filtered_merged_links']:
for node in link.out:
if node.t == types.ConceptNode and node != self.ret[0]:
name += node.name + ", "
log.debug(name + ": " + str(result['interaction_information']))
def __make_result_log(self, result_list, reverse):
# To print(debug) interaction information value.
if reverse:
result_list = reversed(result_list)
for i, result in enumerate(result_list):
name = ""
# Prints only top 5 results.
if i < 15:
for link in result['filtered_merged_links']:
for node in link.out:
if node.t == types.ConceptNode and node != self.ret[0]:
name += node.name + ", "
log.debug(name + ": " + str(result['interaction_information']))
def __evaluate_interaction_information(
self,
decided_atoms,
conflict_link_cases,
non_conflict_link_cases,
non_duplicate_link_cases,
):
"""Evaluate interaction information value for each available conflict
link cases, and returns one link set has maximum information value.
Args:
decided_atoms: The source atoms to make new atom.
conflict_link_cases: Conflicted link tuples list.
non_conflict_link_cases: Non-conflict links list.
non_duplicate_link_cases: Non-duplicated links list.
:param decided_atoms: list[Atom]
:param conflict_link_cases: list[list[EqualLinkKey]]
:param non_conflict_link_cases: list[EqualLinkKey]
:param non_duplicate_link_cases: list[EqualLinkKey]
Returns:
A link set has maximum interaction information value.
:rtype: list[EqualLinkKey]
"""
result_list = list()
inheritance_nodes = list()
for decided_atom in decided_atoms:
inheritance_nodes += find_inheritance_nodes(self.a, decided_atom)
# TODO: To prevent freeze during interaction information generation,
# user can limit the max value of calculation.
max_repeat_length = self.evaluate_n_gram_limit \
if self.evaluate_n_gram_limit < self.provider.n_gram \
else self.provider.n_gram
log.debug("ConnectConflictInteractionInformation: " +
"Calculating interaction information " + "(Total: " +
str(len(conflict_link_cases)) + ")")
current_ratio = 0
for i, conflict_link in enumerate(conflict_link_cases):
current_ratio = self.__print_progress(
"ConnectConflictInteractionInformation:II:", current_ratio, i,
len(conflict_link_cases), 25)
merged_links = list()
merged_links.extend(non_conflict_link_cases)
merged_links.extend(non_duplicate_link_cases)
merged_links.extend(conflict_link)
# Calculate n-gram data in each available link cases.
# The provider with high n-gram will provides more correct data,
# but speed will going slower rapidly.
filtered_merged_links = \
map(lambda x:
eq_link.key_to_link(self.a, x, self.ret[0], x.tv),
filter(lambda x:
# TODO: Currently connector excludes
# an InheritanceLink to get valuable(funny) result.
self.a[x.h].out[0] not in inheritance_nodes and
# Manually throw away the links have low strength.
(x.tv.mean > self.inter_info_strength_above_limit),
merged_links
)
)
if len(filtered_merged_links) is 0:
continue
interaction_information = PyInteractionInformationAtom(). \
calculate_interaction_information(
filtered_merged_links,
self.provider,
max_repeat_length
)
result_list.append({
"merged_links":
merged_links,
"filtered_merged_links":
filtered_merged_links,
"interaction_information":
interaction_information
})
if len(result_list) < 1:
self.last_status = blending_status.EMPTY_RESULT
return []
result_list = sorted(result_list,
key=(lambda x: x["interaction_information"]),
reverse=True)
self.__make_result_log(result_list, reverse=False)
# self.__make_result_log(result_list, reverse=True)
return result_list[0]['merged_links']
def __evaluate_interaction_information(
self,
decided_atoms,
conflict_link_cases,
non_conflict_link_cases,
non_duplicate_link_cases,
):
"""Evaluate interaction information value for each available conflict
link cases, and returns one link set has maximum information value.
Args:
decided_atoms: The source atoms to make new atom.
conflict_link_cases: Conflicted link tuples list.
non_conflict_link_cases: Non-conflict links list.
non_duplicate_link_cases: Non-duplicated links list.
:param decided_atoms: list[Atom]
:param conflict_link_cases: list[list[EqualLinkKey]]
:param non_conflict_link_cases: list[EqualLinkKey]
:param non_duplicate_link_cases: list[EqualLinkKey]
Returns:
A link set has maximum interaction information value.
:rtype: list[EqualLinkKey]
"""
result_list = list()
inheritance_nodes = list()
for decided_atom in decided_atoms:
inheritance_nodes += find_inheritance_nodes(self.a, decided_atom)
# TODO: To prevent freeze during interaction information generation,
# user can limit the max value of calculation.
max_repeat_length = self.evaluate_n_gram_limit \
if self.evaluate_n_gram_limit < self.provider.n_gram \
else self.provider.n_gram
log.debug(
"ConnectConflictInteractionInformation: " +
"Calculating interaction information " +
"(Total: " + str(len(conflict_link_cases)) + ")"
)
current_ratio = 0
for i, conflict_link in enumerate(conflict_link_cases):
current_ratio = self.__print_progress(
"ConnectConflictInteractionInformation:II:",
current_ratio, i, len(conflict_link_cases), 25
)
merged_links = list()
merged_links.extend(non_conflict_link_cases)
merged_links.extend(non_duplicate_link_cases)
merged_links.extend(conflict_link)
# Calculate n-gram data in each available link cases.
# The provider with high n-gram will provides more correct data,
# but speed will going slower rapidly.
filtered_merged_links = \
map(lambda x:
eq_link.key_to_link(self.a, x, self.ret[0], x.tv),
filter(lambda x:
# TODO: Currently connector excludes
# an InheritanceLink to get valuable(funny) result.
self.a[x.h].out[0] not in inheritance_nodes and
# Manually throw away the links have low strength.
(x.tv.mean > self.inter_info_strength_above_limit),
merged_links
)
)
if len(filtered_merged_links) is 0:
continue
interaction_information = PyInteractionInformationAtom(). \
calculate_interaction_information(
filtered_merged_links,
self.provider,
max_repeat_length
)
result_list.append({
"merged_links": merged_links,
"filtered_merged_links": filtered_merged_links,
"interaction_information": interaction_information
})
if len(result_list) < 1:
self.last_status = blending_status.EMPTY_RESULT
return []
result_list = sorted(
result_list,
key=(lambda x: x["interaction_information"]),
reverse=True
)
self.__make_result_log(result_list, reverse=False)
# self.__make_result_log(result_list, reverse=True)
return result_list[0]['merged_links']
