def test_run_no_answer_sets(self):
facts = [
literal.Literal("person", ["patrick"]),
literal.Literal("person", ["patrick"], positive=False)
]
result = self.solver.run(self.ontology, facts)
self.assertEqual([], result)
def test_run_multiple_answer_sets(self):
facts = [literal.Literal("person", ["patrick"])]
target_1 = answer_set.AnswerSet(facts, [literal.Literal("hero", ["patrick"])])
target_2 = answer_set.AnswerSet(facts, [literal.Literal("hero", ["patrick"], positive=False)])
result = self.solver.run(self.ontology, facts)
self.assertEqual(2, len(result))
self.assertTrue(
(target_1 == result[0] and target_2 == result[1]) or
(target_1 == result[1] and target_2 == result[0])
)
def test_eq(self):
# CHECK: equality checks work as expected
self.assertTrue(
answer_set.AnswerSet([literal.Literal("fact-1"), literal.Literal("fact-2")], []) ==
answer_set.AnswerSet([literal.Literal("fact-2"), literal.Literal("fact-1")], [])
)
self.assertFalse(
answer_set.AnswerSet([literal.Literal("fact-1")], [literal.Literal("fact-2")]) ==
answer_set.AnswerSet([literal.Literal("fact-2")], [literal.Literal("fact-1")])
)
def run(
self, path: str, facts: typing.Iterable[literal.Literal]
) -> typing.List[answer_set.AnswerSet]:
# sanitize args
path = str(path)
if not os.path.isfile(path):
raise ValueError(
"The provided <path> does not refer to an existing file: '{}'!"
.format(path))
insanity.sanitize_type("facts", facts, collections.Iterable)
facts = set(facts)
insanity.sanitize_iterable("facts",
facts,
elements_type=literal.Literal)
self._sanitize_literals(facts)
# prepare facts as single string to provide to DLV
str_facts = ". ".join(str(f) for f in facts)
if str_facts:
str_facts += "."
# run DLV
cmd = "echo \"{}\" | {} -silent -- {}".format(str_facts,
self._dlv_path, path)
result = str(
subprocess.check_output(cmd, shell=True,
universal_newlines=True)).strip()
# check if any answer set has been provided at all
if result == "":
return []
# split result into parts representing single answer sets
result = [r.strip()[1:-1] for r in result.split("\n")]
# create answer sets
result_sets = []
for r in result: # iterate over all answer sets (i.e., string representations of them)
# collect inferences
inferences = set()
if r != "":
for x in r.split(", "):
m = re.match(self.LITERAL_PATTERN, x)
lit = literal.Literal(m.group("predicate"),
m.group("terms").split(","),
positive=m.group("sign") == "")
if lit not in facts:
inferences.add(lit)
# create answer set
result_sets.append(answer_set.AnswerSet(facts, inferences))
return result_sets
def test_eq(self):
# CHECK: equality checks work as expected
self.assertTrue(
literal.Literal("person", ["patrick"]) == literal.Literal(
"person", ["patrick"]))
self.assertTrue(
literal.Literal("person", positive=False) == literal.Literal(
"person", positive=False))
self.assertFalse(
literal.Literal("person", ["patrick"]) == literal.Literal(
"person"))
self.assertFalse(
literal.Literal("person", positive=False) == literal.Literal(
"person", positive=True))
def test_run(self):
# CHECK: providing a non-existing ontology path causes a ValueError
with self.assertRaises(ValueError):
self.solver.run("/not/a/valid/path", [])
with self.assertRaises(ValueError):
self.solver.run(None, [])
# CHECK: providing a non-iterable as facts causes a TypeError
with self.assertRaises(TypeError):
self.solver.run(self.ontology, "no facts")
with self.assertRaises(TypeError):
self.solver.run(self.ontology, None)
with self.assertRaises(TypeError):
self.solver.run(self.ontology, 123)
# CHECK: if facts contains instances of a type other than Literal, then a TypeError is raised
with self.assertRaises(TypeError):
self.solver.run(self.ontology, [literal.Literal("person", ["patrick"]), 0])
with self.assertRaises(TypeError):
self.solver.run(self.ontology, [None, literal.Literal("person", ["patrick"])])
# CHECK: correct invocations of run do not causes any issues
self.solver.run(self.ontology, [])
self.solver.run(self.ontology, (literal.Literal("person", ["patrick"]), literal.Literal("hero", ["patrick"])))
def test_iter(self):
# CHECK: iterating over an answer sets covers all facts and inferences
ans = answer_set.AnswerSet(
[literal.Literal("fact-1"), literal.Literal("fact-2")],
[literal.Literal("fact-3")]
)
self.assertEqual(
{literal.Literal("fact-1"), literal.Literal("fact-2"), literal.Literal("fact-3")},
set(ans)
)
def test_init(self):
# CHECK: the predicate symbol must not be the empty string
with self.assertRaises(ValueError):
literal.Literal("")
# CHECK: none of the terms can be the empty string
with self.assertRaises(ValueError):
literal.Literal("pred", [""])
with self.assertRaises(ValueError):
literal.Literal("pred", ["abc", "def", "", "jkl"])
# CHECK: legal args do not cause any issues
literal.Literal("person")
literal.Literal("person", [], positive=False)
literal.Literal("person", ["patrick"])
literal.Literal(1) # -> translated to "1"
literal.Literal(None) # -> translated to "None"
# CHECK: attributes are created correctly
lit = literal.Literal("pred", ["a", "b", "c"], positive=False)
self.assertEqual("pred", lit.predicate)
self.assertEqual(("a", "b", "c"), lit.terms)
self.assertFalse(lit.positive)
def _generate_sample(
self,
spec_countries: typing.List[str],
inf_countries: typing.List[str]=None,
minimal: bool=False
) -> kg.KnowledgeGraph:
"""Generates a single training samples based on the provided data.
The arg ``spec_countries`` contains the names of all countries that are supposed to be specified, i.e., fully
known, in the dataset to create, and ``inf_countries`` indicates the names of those whose regions are to be
inferred. If ``inf_countries`` is not provided, however, then the prediction targets are chosen randomly from
``spec_countries``.
Args:
spec_countries (list[str]): All countries whose regions are specified.
inf_countries (list[str], optional): Those countries, whose regions are to be inferred.
minimal (bool, optional): Specifies whether to generate a minimal sample, i.e., one that contains inferences
and predictions for target countries only. This is ``False``, by default.
Returns:
kg.KnowledgeGraph: The created training sample.
"""
# randomly shuffle countries
countries = spec_countries[:]
if inf_countries:
countries += inf_countries
random.shuffle(countries)
# (randomly) choose prediction targets if not provided
if inf_countries:
inf_countries = set(inf_countries)
else:
inf_countries = set(countries[-self.NUM_EVAL_COUNTRIES:])
# determine all countries that are neighbors of a prediction target (but not targets by themselves)
inf_neighbors = set()
for c in inf_countries:
inf_neighbors |= set(self._data[c].neighbors)
inf_neighbors -= set(inf_countries)
# create new knowledge graph and add vocabulary
sample = kg.KnowledgeGraph()
sample.classes.add_all(self._classes.values())
sample.relations.add_all(self._relations.values())
# a dict that maps names to individual objects
individuals = {}
# create variables for storing facts
class_facts = set() # all (positive) class memberships (negative ones are inferred from these)
neighbor_facts = set() # all facts about (positive) neighbor-of relations (negative ones are inferred)
location_facts = set() # the part of the (positive) located-in facts to infer the remaining ones from
all_locations = set() # all (positive) located-in relations (negatives ones are inferred from these)
# create individuals for all regions/subregions
for region in itertools.chain(*((r, *s) for r, s in self._regions.items())):
individuals[region] = ind_fac.IndividualFactory.create_individual(region)
sample.individuals.add(individuals[region])
# create literals that describe the existing regions and subregions as well as the relations among them
for r, subregions in self._regions.items():
class_facts.add(literal.Literal(voc.CLASS_REGION, [r]))
for s in subregions:
class_facts.add(literal.Literal(voc.CLASS_SUBREGION, [s]))
loc_lit = literal.Literal(voc.RELATION_LOCATED_IN, [s, r])
location_facts.add(loc_lit)
all_locations.add(loc_lit)
# create individuals for all countries
for c in countries:
individuals[c] = ind_fac.IndividualFactory.create_individual(c)
sample.individuals.add(individuals[c])
# create literals for (countries') located-in and neighbor-of relationships
for cou_name in countries:
# fetch the current country's region and subregion
r = self._data[cou_name].region
s = self._data[cou_name].subregion
# create literals that describe the country as well as the relation to its region/subregion
cou_lit = literal.Literal(voc.CLASS_COUNTRY, [cou_name])
reg_lit = literal.Literal(voc.RELATION_LOCATED_IN, [cou_name, r])
sub_lit = None if s is None else literal.Literal(voc.RELATION_LOCATED_IN, [cou_name, s])
class_facts.add(cou_lit)
all_locations.add(reg_lit)
if sub_lit is not None:
all_locations.add(sub_lit)
# determine whether the located-in predicates should be added to the list of provided facts
if self._problem_setting == self.PROBLEM_S1:
if sub_lit is not None: # subregion is provided for all countries
location_facts.add(sub_lit)
if cou_name not in inf_countries: # region is not provided for target countries
location_facts.add(reg_lit)
elif self._problem_setting == self.PROBLEM_S2:
if cou_name not in inf_countries: # neither region nor subregion are provided for target countries
location_facts.add(reg_lit)
if sub_lit is not None:
location_facts.add(sub_lit)
else:
if cou_name not in inf_countries and cou_name not in inf_neighbors: # region is neither provided for
location_facts.add(reg_lit) # for targets nor their neighbors
if cou_name not in inf_countries and sub_lit is not None: # subregion is not provided for
location_facts.add(sub_lit) # target countries
# iterate over all neighbors of the current country, and add according neighbor-of literals
for n in self._data[cou_name].neighbors:
if n in countries: # -> important, because not all of the countries in self._data might be used
neighbor_facts.add(literal.Literal(self.NEIGHBOR_OF_PREDICATE, [cou_name, n]))
neighbor_facts.add(literal.Literal(self.NEIGHBOR_OF_PREDICATE, [n, cou_name]))
# compute all inferences that are possible based on the restricted data
input_facts = list(itertools.chain(neighbor_facts, location_facts))
if self._class_facts:
input_facts += class_facts
answer_set = self._solver.run(self._ontology_path, input_facts)[0]
# add all facts to the sample
for f in list(sorted(answer_set.facts, key=lambda x: str(x))):
self._add_literal_to_kg(sample, individuals, f)
# add all inferences ot the sample
for i in list(sorted(answer_set.inferences, key=lambda x: str(x))):
if (
not minimal or
i.predicate == "region" or
i.predicate == "subregion" or
(len(i.terms) == 1 and i.terms[0] in inf_countries) or
(len(i.terms) == 2 and (i.terms[0] in inf_countries or i.terms[1] in inf_countries))
):
self._add_literal_to_kg(sample, individuals, i, inferred=True)
# compute perfect knowledge
perfect_knowledge = set(
self._solver.run(
self._ontology_path,
itertools.chain(class_facts, neighbor_facts, all_locations)
)[0]
)
# determine all information that was neither provided nor inferred
missing_knowledge = list(sorted(perfect_knowledge - set(answer_set), key=lambda x: str(x)))
# add missing knowledge as prediction targets to the sample
for p in missing_knowledge:
if (
not minimal or
p.predicate == "region" or
p.predicate == "subregion" or
(len(p.terms) == 1 and p.terms[0] in inf_countries) or
(len(p.terms) == 2 and (p.terms[0] in inf_countries or p.terms[1] in inf_countries))
):
self._add_literal_to_kg(sample, individuals, p, prediction=True)
# provide the created sample
return sample
def test_sanitize_literals(self):
# CHECK: illegal predicate of term symbols cause a ValueError
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("Person", ["patrick"])])
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("1person", ["patrick"])])
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("per-son", ["patrick"])])
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("_person", ["patrick"])])
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("person", ["Patrick"])])
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("person", ["1patrick"])])
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("person", ["pat-rick"])])
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("person", ["_patrick"])])
with self.assertRaises(ValueError):
self.solver._sanitize_literals([literal.Literal("person", ["patrick", "Patrick"])])
# CHECK: legal literals do not cause any issues
self.solver._sanitize_literals([literal.Literal("person", ["patrick"])])
self.solver._sanitize_literals([literal.Literal("person1", ["patrick1"])])
self.solver._sanitize_literals([literal.Literal("per_son", ["pat_rick"])])
self.solver._sanitize_literals([literal.Literal("personPerson", ["patrickPatrick"])])
def test_run_single_answer_sets(self):
facts = [literal.Literal("hero", ["patrick"])]
target = answer_set.AnswerSet(facts, [literal.Literal("person", ["patrick"])])
result = self.solver.run(self.ontology, facts)
self.assertEqual(1, len(result))
self.assertEqual(target, result[0])
def test_init(self):
# CHECK: facts has to be an iterable of literals -> otherwise a TypeError is raised
with self.assertRaises(TypeError):
answer_set.AnswerSet("facts", [])
answer_set.AnswerSet(literal.Literal("fact"), [])
with self.assertRaises(TypeError):
answer_set.AnswerSet(["fact"], [])
with self.assertRaises(TypeError):
answer_set.AnswerSet((literal.Literal("test"), "fact"), [])
# CHECK: inferences has to be an iterable of literals -> otherwise a TypeError is raised
with self.assertRaises(TypeError):
answer_set.AnswerSet([], "facts")
answer_set.AnswerSet([], literal.Literal("fact"))
with self.assertRaises(TypeError):
answer_set.AnswerSet([], ["fact"])
with self.assertRaises(TypeError):
answer_set.AnswerSet([], (literal.Literal("test"), "fact"))
# CHECK: providing legal values causes no issues
answer_set.AnswerSet([], [])
answer_set.AnswerSet([literal.Literal("fact-1")], [])
answer_set.AnswerSet({literal.Literal("fact-1"), literal.Literal("fact-2")}, [])
answer_set.AnswerSet([], [literal.Literal("fact-1")])
answer_set.AnswerSet([], (literal.Literal("fact-1"), literal.Literal("fact-2")))
answer_set.AnswerSet([literal.Literal("fact-1")], [literal.Literal("fact-2")])
# CHECK: attributes are defined correctly
ans = answer_set.AnswerSet(
[literal.Literal("fact-1"), literal.Literal("fact-2")],
(literal.Literal("fact-3"), literal.Literal("fact-4"), literal.Literal("fact-5"))
)
self.assertEqual(
{literal.Literal("fact-1"), literal.Literal("fact-2")},
ans.facts
)
self.assertEqual(
{literal.Literal("fact-3"), literal.Literal("fact-4"), literal.Literal("fact-5")},
ans.inferences
)
