def test_alternative_partial_order(self):
class OrderedPair:
"""Pair (a,b) where (a,b) are natural numbers.
Partial order is defined by (a,b) <= (c,d) iff a<=c and b<=d."""
def __init__(self, x, y):
self.x: int = x
self.y: int = y
def __eq__(self, other):
return self.x == other.x and self.y == other.y
def __le__(self, other):
return self.x <= other.x and self.y <= other.y
def __hash__(self):
return hash((self.x, self.y))
default = Case('default', OrderedPair(0, 0), outcome=0)
case1 = Case('1', OrderedPair(1, 0), outcome=1)
case2 = Case('2', OrderedPair(0, 1), outcome=0)
case3 = Case('3', OrderedPair(2, 1), outcome=0)
cb = (case1, case2, case3)
clf = Aacbr(default_case=default)
clf.fit(cb)
assert set(clf.casebase_active) == set(cb + (default, ))
test = [
OrderedPair(2, 0),
OrderedPair(0, 2),
OrderedPair(20, 20),
OrderedPair(1, 1),
OrderedPair(0, 0)
]
expected_output = [1, 0, 0, 1, 0]
predictions = clf.predict(test)
assert expected_output == predictions
def test_graph_drawing(self, tmp_path):
"Checks if a graph is created"
cb = self.example_cb2
clf = Aacbr().fit(cb)
clf.draw_graph(output_dir=tmp_path)
output_path = tmp_path / "graph.png"
assert output_path.exists()
assert output_path.is_file()
def test_inconsistent(self):
# Even if already tested in test_cases.py, Aacbr should have its
# own interface to it.
case1 = Case('1', {'a', 'b'}, outcome=0)
case2 = Case('2', {'a', 'b'}, outcome=1)
cb = [case1, case2]
clf = Aacbr().fit(cb)
assert clf.past_case_attacks(case1, case2)
assert clf.past_case_attacks(case2, case1)
assert clf.inconsistent_attacks(case1, case2)
assert clf.inconsistent_attacks(case2, case1)
def test_attack(self, cb):
clf = Aacbr()
clf.fit(cb)
if cb == self.example_cb or cb == self.example_cb2:
list_of_attacks = ((self.case1, self.default), (self.case2,
self.case1))
else:
raise (Exception("Undefined test"))
for pair in product(cb, repeat=2):
assert ((clf.past_case_attacks(pair[0], pair[1])) == (
pair in list_of_attacks
)), f"Violated by pair {pair}. Expected {pair in list_of_attacks}."
def test_argumentation_framework(self):
cb = self.example_cb
# newcase = self.case3
newcase = Case('4', {'a', 'd'}, outcome=1)
expected_output = newcase.outcome
clf = Aacbr().fit(cb)
framework = clf.give_argumentation_framework(newcase)
arguments, attacks = framework
assert arguments == set(cb + (newcase, ))
expected_attacks = \
{(self.case2, self.case1),
(self.case1, self.default)}
assert attacks == expected_attacks
def test_scikit_learning_like_api_with_case_input(self):
# It would be nice to have it compatible with the scikit-learn API:
# https://scikit-learn.org/stable/developers/develop.html#apis-of-scikit-learn-objects
train_data = self.example_cb2
test_data = [
Case('new1', {'a'}),
Case('new2', {'a', 'b'}),
Case('new3', {'a', 'c'}),
Case('new4', {'a', 'b', 'c', 'd'}),
Case('new5', set())
]
expected_output = [1, 0, 1, 0, 0]
clf = Aacbr()
predicted_output = clf.fit(train_data).predict(test_data)
assert expected_output == predicted_output
def test_default_case_in_casebase(self):
default = Case('default', {'a'}, outcome=0)
case1 = Case('1', {'a', 'b'}, outcome=0)
case2 = Case('2', {'a', 'b'}, outcome=1)
cb = [default, case1, case2]
clf = Aacbr().fit(cb)
assert clf.default_case == default
def test_default_case_in_arguments(self):
default = Case('default', {'a'}, outcome=0)
case1 = Case('1', {'a', 'b'}, outcome=0)
case2 = Case('2', {'a', 'b'}, outcome=1)
cb = [case1, case2]
clf = Aacbr(default_case=default).fit(cb)
assert clf.default_case == default
def test_attack_new_case(self, cb):
new = Case('new', {'a', 'd'})
clf = Aacbr()
clf.fit(cb)
assert not clf.new_case_attacks(new, self.default)
if self.case1 in cb:
assert not clf.new_case_attacks(new, self.case1)
if self.case2 in cb:
assert clf.new_case_attacks(new, self.case2)
if self.case3 in cb:
assert clf.new_case_attacks(new, self.case3)
def test_scikit_learning_like_api_with_characterisation_input(self):
case0 = Case("0", set(), outcome=0)
cb = [case0, self.case1, self.case2, self.case3]
train_X = [c.factors for c in cb]
train_Y = [c.outcome for c in cb]
test_data = [
Case('new1', {'a'}),
Case('new2', {'a', 'b'}),
Case('new3', {'a', 'c'}),
Case('new4', {'a', 'b', 'c', 'd'}),
Case('new5', set())
]
test_X = [c.factors for c in test_data]
expected_output = [1, 0, 1, 0, 0]
clf = Aacbr()
clf.fit(train_X, train_Y)
assert set(clf.casebase_active) == set(cb + [self.default])
predicted_output = clf.predict(test_X)
assert expected_output == predicted_output
def test_scikit_learning_like_api_with_characterisation_input2(self):
train_X = [set(), {'a'}, {'a', 'b'}, {'a', 'b', 'c'}]
train_Y = [0, 1, 0, 0]
test_X = [{'a'}, {'a', 'b'}, {'a', 'c'}, {'a', 'b', 'c', 'd'}, set()]
expected_output = [1, 0, 1, 0, 0]
clf = Aacbr()
clf.fit(train_X, train_Y)
default = Case('default', set(), outcome=0)
case0 = Case("0", set(), outcome=0)
case1 = Case('1', {'a'}, outcome=1)
case2 = Case('2', {'a', 'b'}, outcome=0)
case3 = Case('3', {'a', 'b', 'c'}, outcome=0)
cb = [case0, case1, case2, case3]
assert set(clf.casebase_active) == set(cb + [default])
predicted_output = clf.predict(test_X)
assert expected_output == predicted_output
def test_inconsistent_IO(self):
default = Case('default', set(), outcome=0)
case1 = Case('1', {'a'}, outcome=1)
case2 = Case('2', {'a'}, outcome=0)
case3 = Case('3', {'a', 'b'}, outcome=0)
case4 = Case('4', {'a', 'b'}, outcome=1)
cb = [default, case1, case2, case3, case4]
train_data = cb
test_data = [
Case('new1', {'a'}),
Case('new2', {'a', 'b'}),
Case('new3', {'a', 'c'}),
Case('new4', {'a', 'b', 'c', 'd'}),
Case('new5', set()),
Case('new6', {'a', 'c', 'd'})
]
expected_output = [1, 1, 1, 1, 0, 1]
clf = Aacbr(cautious=False)
predicted_output = clf.fit(train_data).predict(test_data)
assert expected_output == predicted_output
def test_remove_spikes(self):
default = Case('default', set(), outcome=0)
case1 = Case('1', {'a'}, outcome=1)
case2 = Case('2', {'a', 'b'}, outcome=0)
case3 = Case('3', {'b'}, outcome=0)
case4 = Case('4', {'c'}, outcome=0)
case5 = Case('5', {'a', 'c'}, outcome=1)
case6 = Case('6', {'a', 'b', 'c'}, outcome=0)
cb = (default, case1, case2, case3, case4, case5, case6)
filtered_cb = {default, case1, case2}
clf = Aacbr().fit(cb, remove_spikes=True)
assert set(clf.casebase_active) == filtered_cb
def test_scikit_learning_like_api_with_case_input_cautious(self):
default = Case('default', set(), outcome=0)
case1 = Case('1', {'a'}, outcome=1)
case2 = Case('2', {'a', 'b'}, outcome=0)
case3 = Case('3', {'c'}, outcome=1)
case4 = Case('4', {'c', 'd'}, outcome=0)
case5 = Case('5', {'a', 'b', 'c'}, outcome=1)
cb = [default, case1, case2, case3, case4, case5]
train_data = cb
test_data = [
Case('new1', {'a'}),
Case('new2', {'a', 'b'}),
Case('new3', {'a', 'c'}),
Case('new4', {'a', 'b', 'c', 'd'}),
Case('new5', set()),
Case('new6', {'a', 'c', 'd'})
]
expected_output = [1, 0, 1, 0, 0, 1]
clf = Aacbr(cautious=True)
predicted_output = clf.fit(train_data).predict(test_data)
assert expected_output == predicted_output
assert set(clf.casebase_active) == set(
[default, case1, case2, case3, case4])
#
clf_noncautious = Aacbr(cautious=False)
expected_output = [1, 0, 1, 1, 0, 1]
predicted_output = clf_noncautious.fit(train_data).predict(test_data)
assert expected_output == predicted_output, "Non-cautious is not giving expected result!"
def test_conciseness(self):
cb = self.example_cb2
clf = Aacbr()
clf.fit(cb)
assert clf.past_case_attacks(self.case2, self.case1)
assert not clf.past_case_attacks(
self.case3, self.case1
), "case3 is attacking case1 even if case2 already does so. Violating conciseness."
def run_test_from_files(aacbr_type, test):
# TODO: change this interface in line below in the future
# -- it should not be inside Aacbr
cautious = True if aacbr_type == "cautious" else False
casebase = load_cases(TEST_PATH_PREFIX + test["casebase"])
clf = Aacbr(outcome_def=test["outcomes"]["default"],
outcome_nondef=test["outcomes"]["nondefault"],
outcome_unknown=test["outcomes"]["undecided"],
cautious=cautious)
clf.fit(casebase)
casebase_active_ids = set(
map(lambda x: getattr(x, "id"), clf.casebase_active))
assert set(casebase_active_ids) == set(
test["casebase_expected"][aacbr_type])
for newcase_spec in test["newcases"]:
newcase = Case(id=newcase_spec["id"],
factors=set(newcase_spec["factors"]))
result = clf.predict([newcase])
# prediction = result[1][0]["Prediction"]
prediction = result[0]
assert prediction == newcase_spec["outcome_expected"][
aacbr_type], f"Failed for {newcase_spec}, in type {aacbr_type}" f"Failed on test {test}"
def test_grounded_extension(self):
default = Case('default', set(), outcome=0)
case1 = Case('1', {'a'}, outcome=1)
case2 = Case('2', {'a', 'b'}, outcome=0)
case3 = Case('3', {'a', 'b', 'c'}, outcome=0)
case4 = Case('4', {'c'}, outcome=1)
example_cb = (default, case1, case2, case3, case4)
new = Case('new', {'a', 'c'})
new2 = Case('new2', {'a', 'b'})
ge = {case1, case4, new}
ge2 = {case2, default, new2}
clf = Aacbr()
clf.fit(example_cb)
assert clf.grounded_extension(new_case=new) == ge
assert clf.grounded_extension(new_case=new2) == ge2
def test_default_case_implict(self):
case1 = Case('1', {'a', 'b'}, outcome=0)
case2 = Case('2', {'a', 'b'}, outcome=1)
cb = [case1, case2]
clf = Aacbr().fit(cb)
assert clf.default_case == Case("default", set(), clf.outcome_def)
def setup(request):
test = request.param
setup_result = Aacbr(test["outcomes"]["default"],
test["outcomes"]["nondefault"],
test["outcomes"]["undecided"])
return test, setup_result
def test_aacbr_methods(self, cb):
clf = Aacbr().fit(cb)
assert clf.casebase_initial == cb
def test_initialisation(self, cb):
clf = Aacbr().fit(cb)
assert isinstance(clf, Aacbr)