def learn(self):
self.running_time = time()
current_attributes = []
while len(current_attributes) < len(self.attributes):
while True:
current_closure = aclosure(current_attributes, self.context)
if current_attributes == current_closure:
break
implication = Implication(set(current_attributes),
set(current_closure))
counterexample = (self.implication_expert.
provide_counterexample(implication))
if counterexample is None:
self.basis.append(implication)
break
else:
self.context.add_object(
_get_mask(sorted(counterexample), self.attributes),
self.context.get_size()[0])
attributes_closure = lambda attributes_subset: sorted(
lin_closure(set(attributes_subset), self.basis))
current_attributes = next_closure(current_attributes,
self.attributes,
attributes_closure)
self.running_time = time() - self.running_time
self.learning_statistics['running time'] = self.running_time
self.learning_statistics['implication queries'] = (
self.implication_expert.implication_query_count)
self.implication_expert.implication_query_count = 0
def setUp(self):
ct = [[True, False, False, True], [True, False, True, False],
[False, True, True, False], [False, True, True, True]]
objs = ['1', '2', '3', '4']
attrs = ['a', 'b', 'c', 'd']
self.cxt = fca.Context(ct, objs, attrs)
self.basis = [Implication({'c', 'd'}, {'b'})]
def setUp(self):
ct = [[True, False, False, True],\
[True, False, True, False],\
[False, True, True, False],\
[False, True, True, True]]
objs = ['1', '2', '3', '4']
attrs = ['a', 'b', 'c', 'd']
self.cxt = fca.Context(ct, objs, attrs)
self.basis = [Implication(set(['c', 'd']), set(['b']))]
def is_member(self, attribute_subset):
""" Checks whether the given subset of attributes is a model of actual
implication basis.
Args:
attribute_subset: An object represented as a subset of
attributes.
Returns:
True if the given subset of attributes is a model of
actual implication basis and False otherwise.
"""
self.membership_query_count += 1
implication = Implication(attribute_subset, set())
for attribute in self.attributes:
if attribute not in attribute_subset:
implication = Implication(attribute_subset, {attribute})
if self.implication_expert.is_valid(implication):
return False
return True
def learn(self, max_iterations=None, verbose=False):
""" Computes PAC-basis by interacting with MembershipEquivalenceExpert.
Args:
max_iterations: The maximum number of iterations for correction
of the hypothetical basis. If None, then the
number of iterations is not limited.
verbose (default=False): Enable verbose output.
"""
self._verbose_learning = verbose
self.running_time = time()
self.pac_basis = []
while True:
self._log_info(
'%d. Hypothetical basis:' %
(self.expert.approximately_equivalence_query_count + 1),
self.pac_basis)
counterexample = self.query_equivalence_counterexample()
if counterexample is None:
return 0
if _closed(counterexample, self.pac_basis):
# Negative counterexample.
self._log_info('Negative counterexample:',
list(counterexample),
end='\n\n')
found = False
for implication in self.pac_basis:
premise = implication.premise
new_premise = premise & counterexample
if new_premise < premise and not self.expert.is_member(
new_premise):
implication.set_premise(new_premise)
found = True
break
if not found:
self.pac_basis.append(
Implication(counterexample, set(self.attributes)))
else:
# Positive counterexample.
self._log_info('Positive counterexample:',
list(counterexample),
end='\n\n')
for implication in self.pac_basis:
if not implication.is_respected(counterexample):
implication.set_conclusion(implication.conclusion
& counterexample)
if (max_iterations
and self.expert.approximately_equivalence_query_count >
max_iterations):
return 1
def updated_basis(intent, basis, attributes):
valid = []
invalid = []
for imp in basis:
if not imp.premise <= intent or imp.conclusion <= intent:
valid.append(imp)
else:
invalid.append(imp)
new_valid = []
for imp in invalid:
new_imp = Implication(imp.premise, imp.conclusion & intent)
add_smartly(new_imp, new_valid, valid)
valid += new_valid
new_valid = []
for imp in invalid:
for a in attributes - intent:
aset = set([a])
new_imp = Implication(imp.premise | aset, imp.conclusion | aset)
if (remove_subsumed(new_imp, valid)
and remove_subsumed_plus(new_imp, new_valid)):
new_valid.append(new_imp)
return valid + new_valid
def compute_implication_cover(cxt, close=closure_operators.closure):
"""
Compute an implication cover for a given *cxt* using
an object-incremental algorithm
"""
attributes = set(cxt.attributes)
basis = [Implication(set(), attributes.copy())]
i = 0
for intent in cxt.examples():
i += 1
print 'object ', i
# print_basis(basis)
# print 'Adding ', intent
# raw_input()
basis = updated_basis(intent, basis, attributes)
print len(basis), 'implications'
return basis
def generalized_compute_dg_basis(attributes,
aclose,
close=closure_operators.simple_closure,
imp_basis=[],
cond=lambda x: True):
"""Compute the Duquenne-Guigues basis using optimized Ganter's algorithm.
*aclose* is a closure operator on the set of attributes.
We need this to implement the exploration
algorithm with partially given examples.
"""
relative_basis = []
a = close(set(), imp_basis)
i = len(attributes)
while len(a) < len(attributes):
a_closed = set(aclose(a))
if a != a_closed and cond(a):
relative_basis.append(Implication(a.copy(), a_closed.copy()))
if (a_closed - a) & set(attributes[:i]):
a -= set(attributes[i:])
else:
if len(a_closed) == len(attributes):
return relative_basis
a = a_closed
i = len(attributes)
for j in range(i - 1, -1, -1):
m = attributes[j]
if m in a:
a.remove(m)
else:
b = close(a | set([m]), relative_basis + imp_basis)
if not (b - a) & set(attributes[:j]):
a = b
i = j
break
return relative_basis
def is_strongly_approximately_equivalent(self, hypothetical_basis, epsilon,
delta):
""" Checks whether the given basis is equivalent to the actual
one with accuracy epsilon and confidence delta.
This function tries to find a counterexample for hypothetical basis
by checking random subsets of attributes generated at random
uniformly.
Args:
hypothetical_basis: A basis to check.
epsilon: The number from (0, 1) indicating the accuracy.
delta: The number from (0, 1) indicating the confidence level.
Returns:
A counterexample if found, None otherwise.
"""
self.approximately_equivalence_query_count += 1
n = self.approximately_equivalence_query_count
q = self.degree
if self.mode == 'telescoping':
num_attempts = math.log(n**q * (n + 1)**q /
(delta * ((n + 1)**q - n**q))) / epsilon
elif self.mode == 'pow2':
num_attempts = (self.approximately_equivalence_query_count -
math.log(delta)) / epsilon
num_attempts = int(math.ceil(num_attempts))
for _ in xrange(num_attempts):
random_subset = _select_random_uniform_subset(self.attributes)
closed_random_subset = lin_closure(random_subset,
hypothetical_basis)
implication = Implication(random_subset, closed_random_subset)
counterexample = self.implication_expert.provide_counterexample(
implication)
if counterexample:
return counterexample
if not self.is_member(closed_random_subset):
return closed_random_subset
return None
def generalized_dg_basis_iter(attributes,
aclose,
close=closure_operators.simple_closure,
imp_basis=[],
cond=lambda x: True):
"""Compute iterator over Duquenne-Guigues basis using optimized Ganter's
algorithm.
*aclose* is a closure operator on the set of attributes.
"""
relative_basis = []
a = close(set(), imp_basis)
i = len(attributes)
while len(a) < len(attributes):
a_closed = set(aclose(a))
if a != a_closed and cond(a):
basis_imp = Implication(a.copy(), a_closed.copy())
yield basis_imp
relative_basis.append(basis_imp)
if (a_closed - a) & set(attributes[:i]):
a -= set(attributes[i:])
else:
if len(a_closed) == len(attributes):
break
a = a_closed
i = len(attributes)
for j in range(i - 1, -1, -1):
m = attributes[j]
if m in a:
a.remove(m)
else:
b = close(a | {m}, relative_basis + imp_basis)
if not (b - a) & set(attributes[:j]):
a = b
i = j
break
# 'Ansett Australia', 'The Australian Airlines Group',
# 'British Midland', 'Lufthansa', 'Mexicana',
# 'Scandinavian Airlines', 'Singapore Airlines',
# 'Thai Airways International', 'United Airlines',
# 'VARIG']
# attributes = ['Latin America', 'Europe', 'Canada', 'Asia Pasific',
# 'Middle East', 'Africa', 'Mexico', 'Carribean',
# 'United States']
# table = [[True, True, True, True, True, False, True, True, True],
# [False, True, False, True, False, False, False, False, True],
# [False, True, False, True, False, False, False, False, True],
# [False, False, False, True, False, False, False, False, False],
# [False, True, True, True, True, True, False, False, True],
# [False, True, False, False, False, False, False, False, False],
# [True, True, True, True ,True, True, True, False, True],
# [True, False, True, False, False, False, True, True, True],
# [True, True, False, True, False, True, False, False, True],
# [False, True, True, True, True, True, False, False, True],
# [True, True, False, True, False, False, False, True, True],
# [True, True, True, True, False, False, True, True, True],
# [True, True, False, True, False, True, True, False, True]]
# cxt = fca.Context(table, objects, attributes)
ct = [[True]]
objs = ['1']
attrs = ['a']
cxt = fca.Context(ct, objs, attrs)
imp_basis = compute_dg_basis(cxt, imp_basis=[Implication(set(), {'a'})])
for imp in imp_basis:
print(imp)
def test_attribute_implications(self):
self.attribute_implications = [
Implication(set([
'Carribean',
]), set([
'United States',
'Carribean',
'Latin America',
])),
Implication(set([
'Mexico',
]), set([
'United States',
'Latin America',
'Mexico',
])),
Implication(
set([
'Africa',
]), set([
'Europe',
'United States',
'Asia Pasific',
'Africa',
])),
Implication(
set([
'Middle East',
]),
set([
'Canada',
'Europe',
'Asia Pasific',
'United States',
'Middle East',
])),
Implication(set([
'United States',
'Asia Pasific',
]), set([
'Europe',
'United States',
'Asia Pasific',
])),
Implication(set([
'Canada',
]), set([
'Canada',
'United States',
])),
Implication(set([
'Europe',
'United States',
]), set([
'Europe',
'United States',
'Asia Pasific',
])),
Implication(set([
'Europe',
'Asia Pasific',
]), set([
'Europe',
'United States',
'Asia Pasific',
])),
Implication(
set([
'Canada',
'Europe',
'Asia Pasific',
'Africa',
'United States',
]),
set([
'Canada',
'Europe',
'Asia Pasific',
'Africa',
'United States',
'Middle East',
])),
Implication(set([
'Latin America',
]), set([
'United States',
'Latin America',
])),
Implication(
set([
'United States',
'Carribean',
'Latin America',
'Mexico',
]),
set([
'Canada',
'United States',
'Carribean',
'Latin America',
'Mexico',
])),
Implication(
set([
'Canada',
'United States',
'Latin America',
]),
set([
'Canada',
'United States',
'Latin America',
'Mexico',
])),
Implication(
set([
'Europe',
'Asia Pasific',
'Africa',
'United States',
'Carribean',
'Latin America',
]),
set([
'Canada',
'Europe',
'Asia Pasific',
'Mexico',
'Africa',
'United States',
'Middle East',
'Carribean',
'Latin America',
])),
]
imp_basis = compute_dg_basis(self.cxt)
self.assertEqual(len(imp_basis), len(self.attribute_implications))
for imp in self.attribute_implications:
self.assertTrue(imp in imp_basis)
def test_relative_basis(self):
imp_basis = [Implication(set(), set(['c']))]
relative_basis = compute_dg_basis(self.cxt, imp_basis=imp_basis)
self.assertFalse(imp_basis[0] in relative_basis)
# 'British Midland', 'Lufthansa', 'Mexicana',
# 'Scandinavian Airlines', 'Singapore Airlines',
# 'Thai Airways International', 'United Airlines',
# 'VARIG']
# attributes = ['Latin America', 'Europe', 'Canada', 'Asia Pasific',
# 'Middle East', 'Africa', 'Mexico', 'Carribean',
# 'United States']
# table = [[True, True, True, True, True, False, True, True, True],
# [False, True, False, True, False, False, False, False, True],
# [False, True, False, True, False, False, False, False, True],
# [False, False, False, True, False, False, False, False, False],
# [False, True, True, True, True, True, False, False, True],
# [False, True, False, False, False, False, False, False, False],
# [True, True, True, True ,True, True, True, False, True],
# [True, False, True, False, False, False, True, True, True],
# [True, True, False, True, False, True, False, False, True],
# [False, True, True, True, True, True, False, False, True],
# [True, True, False, True, False, False, False, True, True],
# [True, True, True, True, False, False, True, True, True],
# [True, True, False, True, False, True, True, False, True]]
# cxt = fca.Context(table, objects, attributes)
ct = [[True]]
objs = ['1']
attrs = ['a']
cxt = fca.Context(ct, objs, attrs)
imp_basis = compute_dg_basis(cxt,
imp_basis=[Implication(set(), set(['a']))])
for imp in imp_basis:
print imp